JP6332444B2 - Image processing device - Google Patents

Description

  The present invention relates to an image processing apparatus that removes noise superimposed on a radiographic image, and more particularly to an image processing apparatus that improves the image quality of each frame constituting a live image.

  A medical institution is equipped with a radiation imaging apparatus that acquires an image of radiation. Some of such radiation imaging apparatuses can continuously capture radiation images and output the results as moving images. Such a moving image is sometimes called a live image (see, for example, Patent Document 2).

  Since a live image is obtained by irradiating a subject with a low-dose radiation, the S / N ratio is worse than an image obtained by still image shooting (spot shooting) and includes a lot of noise. It is out. The radiation imaging apparatus includes an image processing apparatus that reduces this noise. This image processing apparatus is configured to obtain a noise-reduced image with an improved S / N ratio by superimposing a plurality of continuous frames over time.

  A recursive filter is a method for generating a noise-reduced image. According to the recursive filter, a noise-reduced image formed by superimposing frames of live images is stored, and this is used as a superposition target image. Then, when a new live image frame is input, a new noise reduced image is generated by superimposing the new live image and the image to be overlapped, and this is stored again. Thereafter, each time a new frame is input, the stored noise-reduced images are overlapped, and noise-reduced images corresponding to the frames are generated one after another. Such an operation of the recursive filter is called a cyclic addition process.

  By the way, the live image is a moving image capturing the movement of the subject. Therefore, by simply superimposing the images, a noise-reduced image in which the images of the subject are doubled is generated. This is because the position and shape at which the subject appears between the two images are different from each other. In addition, since the degree of deviation between the subject images varies depending on the portion of the image, duplication of the subject image cannot be prevented simply by shifting one image in a certain direction and superimposing it on the other image. . The subject image on the live image has a portion where the image matches and a portion where the image does not match between the frames. For the matching part, noise can be reduced by simply superimposing the images. However, the subject image will be blurred in a portion that does not match.

  Therefore, according to the conventional configuration, a configuration that prevents duplication of the image by adopting various devices is adopted. For example, according to Patent Document 1, the superposition mode is changed for each pixel constituting the image. That is, for a portion where the pixel value changes drastically between the frame and the image to be overlapped (a portion where the image does not match), processing is performed so that the pixel value of the frame is inherited as it is. This is because, if simple superposition is performed at this portion, there is a high possibility that the image will be doubled. For a portion that does not change much between images (a portion where the images match), processing is performed such that the frame and the image to be overlapped are superimposed.

  Japanese Patent Laid-Open No. 2004-228561 has a configuration in which the superposition mode is changed for each pixel in consideration of not only the difference in pixel values seen between the images but also the amount of noise reflected in the images. By doing in this way, the part where a lot of noise in the frame appears is selectively subjected to noise reduction processing. Thereby, suppression of duplication of the image and removal of noise are realized at the same time.

  And in patent document 3, it is the structure which investigates where each pixel on a flame | frame corresponds on the superimposition object image by image analysis, and superimposes the pixel on a flame | frame and the corresponding pixel on superimposition object image, It has become. With such a configuration, the structure on the frame and the structure on the polymerization target image are overlapped while being aligned. According to this configuration, not only duplication of the image can be prevented, but noise should be surely removed.

JP 07-154651 A JP 07-079956 JP 01-314477

However, the conventional configuration as described above has the following problems.
That is, with the conventional configuration, it is not possible to reliably overlay images.

  According to the configurations of Patent Documents 1 and 2, noise is hardly reduced at a portion where duplication of an image occurs due to superposition on a frame. This is because the processing is given up because the image is duplicated when the noise reduction processing is performed. Therefore, the methods of these documents cannot completely remove noise on the frame.

  Also, the configuration of Patent Document 3 seems to have no problem at first glance. However, in the method of this document, the pixel of the image to be overlapped corresponding to the pixel on the frame is searched by comparing the pixel of the image to be overlapped one by one with respect to one pixel on the frame. Therefore, it is often the case that the corresponding pixel is erroneously recognized. In such a situation, the images cannot be correctly superimposed and a highly visible noise-reduced image cannot be obtained.

  The present invention has been made in view of such circumstances, and an object of the present invention is to provide an image processing apparatus capable of reliably removing noise of each frame constituting a live image.

The present invention has the following configuration in order to solve the above-described problems.
That is, the image processing apparatus according to the present invention is configured to form an original image in which an object image is reflected in an image processing apparatus that performs noise reduction processing on an image generated by continuously capturing an image of the object. Target setting means for setting a target block composed of a pixel of interest and peripheral pixels surrounding the pixel of interest, and a target block from a reference image in which a subject photographed at a time different from the original image is reflected And a search means for finding a destination block most similar to the target block on the original image, the target block on the original image in the target image on which the subject image is reflected at the same position as the reference image. Overlapping means to generate fusion blocks by superimposing, and target blocks are set one after another while changing the position of the target pixel The image generation means generates a noise-reduced image in which the noise reflected in the original image is reduced by superimposing the fusion blocks on the image one after another. The fusion block is operated so that the position of the fusion block is the same as the position of the target block on the original image.

[Operation / Effect] According to the present invention, it is possible to provide an image processing apparatus capable of reliably removing noise of each frame constituting a live image. That is, according to the present invention, a target block is set on the original image, and a search is made for where the target block is reflected in the superimposition target image. In this way, if the transfer destination of the target block is searched from among the images to be polymerized, it is possible to superimpose the images while tracking the object image on the original image on the image to be polymerized. The subject does not appear twice.
In addition, according to the present invention, since the subject image is tracked in units of blocks, the reliability is remarkably increased as compared with the conventional method of searching for the movement destination for each pixel.

  In the above-described image processing apparatus, the image generation unit adds the fusion blocks while overlapping the fusion blocks, and then adds the pixel values of the pixels on the image to indicate how many times the fusion blocks are added in the pixel. It is more desirable to generate a noise-reduced image by dividing by the number of times.

[Operation / Effect] The above-described configuration explains a more specific configuration of the present invention. In other words, after adding the fusion blocks while overlapping, the pixel value of the pixel on the image is divided by the number of integrations indicating how many times the fusion block is added in that pixel. High noise reduction images can be generated.
Each pixel value of the pixels constituting the noise reduced image is a result of superimposing the target block and the block to be overlapped. The overlap target block should be the same as the subject image reflected in the target block, but if the search means misidentifies the destination block, it is appropriate for the overlap target block in the overlap target image. There are many cases where no part is selected as a polymerization target block. According to the above configuration, each pixel constituting the noise reduced image is not configured based on one overlap target block. That is, each pixel is configured by overlapping different overlapping blocks to be overlapped. Therefore, even if the recognition target block is erroneously recognized, the influence on the noise-reduced image is limited. This is because the polymerization target block related to erroneous recognition is only one of the multiple polymerization target blocks that are overlapped.
According to the above configuration, after adding the fusion blocks, the pixel value of each pixel is divided by the number of integrations of the fusion block to generate a noise reduced image. Therefore, the pixel value level of the noise reduced image is the same as that of the original image. It will be almost the same.

  In the above-described image processing apparatus, the superimposing unit superimposes the pixel of the target block on the target image corresponding to the target block pixel on the original image with the individual weight for each pixel constituting the fusion block. As the absolute value of the difference between the pixel value of the pixel belonging to the target block and the pixel value of the corresponding pixel on the overlap target block or destination block increases, the overlap target block gradually inherits to the fusion block. It is more desirable to change the weight of the overlay so that it is not.

  [Operation / Effect] The above-described configuration explains a more specific configuration of the present invention. The superimposing means superimposes the pixel of the target block on the original image and the pixel of the target block on the corresponding target image on the original image with individual weighting for each pixel constituting the fusion block, and the pixels belonging to the target block If the absolute value of the difference between the pixel value of and the pixel value of the overlap target block or the corresponding pixel on the destination block increases, the overlap weight is gradually changed so that the overlap target block is not inherited by the fusion block. In the original image, the overlapping of the images to be superposed is slight for the portion where the movement of the subject image is intense. In this way, it is possible to more reliably prevent the subject image from being duplicated in the noise reduced image.

  In the above-described image processing apparatus, the superimposing unit is configured to superimpose the target block on the original image and the target block on the target image with individual weighting every time a fusion block is generated. As the absolute value of the difference between the pixel value on the block and the pixel value on the block to be overlapped or the destination block increases, it is more necessary to change the overlay weight so that the block to be overlapped is not succeeded to the fusion block. desirable.

  [Operation / Effect] The above-described configuration explains a more specific configuration of the present invention. In the method of changing the weight for each pixel, the image processing takes too much time. Therefore, as the absolute value of the difference between the pixel value on the target block and the pixel value on the overlap target block or destination block increases, the overlay weight is changed so that the overlap target block is not inherited by the fusion block. In this case, since the weight change is performed in units of blocks, it is possible to perform image processing at a higher speed.

  Further, in the above-described image processing apparatus, when the same position block at the position of the target block is set in the reference image and the movement destination block is not similar to the target block as compared with the same position block, search means It is more desirable to have an editing means for performing overwriting on the output of the search means so that the destination block found by is set to the same position block.

[Operation / Effect] The above-described configuration explains a more specific configuration of the present invention. In both the target block on the original image and the block at the same position on the reference image, if the subject image reflected in both blocks does not move or if the subject image is not reflected in both blocks, The same position block on the reference image is in a state where it is guaranteed to be somewhat similar to the target block. Therefore, when generating a fusion block, it is more visible to superimpose the part corresponding to the target block in the image to be overlapped as it is than to search and superimpose the block corresponding to the target block from the image to be overlapped by the search means. May improve.
According to the above configuration, in the case of the predetermined condition, the search result by the search unit is discarded, and the portion corresponding to the target block in the image to be overlapped is directly overlapped. The blocks to be superposed that have been certified to have moved although they were not originally moved are not overlapped to generate a fusion block, and the visibility of the noise-reduced image is improved.

  In the above-described image processing apparatus, (a1) the superimposition target image is a noise reduction image obtained when image processing is performed on an image photographed before the original image, and (b1) the reference image is the original image. It can be an image taken before.

  In the above-described image processing apparatus, (a1) the superposition target image is a noise reduction image obtained when image processing is performed on an image photographed before the original image, and (b2) the reference image is the original image. It can be set as the noise reduction image corresponding to the image image | photographed previously.

  In the above-described image processing apparatus, (a2) the superimposition target image can be an image photographed before the original image, and (b1) the reference image can be an image photographed before the original image.

  In the above-described image processing apparatus, (a2) the superposition target image is an image photographed before the original image, and (b2) the reference image is a noise-reduced image corresponding to the image photographed before the original image. can do.

  [Operation / Effect] As described above, the image processing apparatus of the present invention can be realized by selecting various modes. Such high selectivity contributes to increasing the degree of freedom of image processing.

  Further, in the above-described image processing apparatus, the search means accurately determines the movement destination block search for a certain target pixel over a wide range of the reference image, and based on the accuracy priority mode search result. The priority operation mode is based on two modes of the speed priority mode for searching for a destination block for a target pixel different from the target pixel that has been processed in the sex priority mode over a narrow range of the reference image. Based on the positional relationship between the target pixel to be processed in the accuracy priority mode and the destination pixel on the reference image of the target pixel found in the accuracy priority mode search, If it operates to predict the destination position of the target pixel on the reference image and search for the destination block over the range surrounding the predicted position Ri desirable.

  [Operation / Effect] The above-described configuration explains a more specific configuration of the present invention. According to the above-described configuration, the search unit performs a search on limited pixels of interest on the original image in an accurate but time-consuming accuracy priority mode. The center of the destination block found by this mode should accurately represent the destination of the target pixel. Therefore, where the pixel around the target pixel in the original image is located on the reference image, it must be around the destination pixel on the reference image of the target pixel. Therefore, when searching for such peripheral pixels, only the periphery of the destination pixel is searched. This is the speed priority mode. With such a configuration, the search means can perform a search operation having both high speed and accuracy.

  In the above-described image processing apparatus, the range of the block to be overlapped according to the overlapping unit may be narrower than the range of the target block and the destination block.

  [Operation / Effect] According to the above-described configuration, it is possible to determine the block to be overlapped by evaluating the region outside the block to be overlapped, so that the fusion block F can be generated with higher reliability. it can.

  In the above-described image processing apparatus, the range of the block to be overlapped according to the overlapping unit may be wider than the range of the target block and the destination block.

  [Operation / Effect] By performing such an operation, it is possible to generate a noise-reduced image from which noise is further removed. This is because if the fusion block F is made larger, the number of pixels superimposed in a multiple manner when a noise-reduced image is generated increases.

  Further, in the above-described image processing apparatus, image reduction means for reducing the original image and the reference image to generate the reduced original image and the reduced reference image, and the target pixel and the reduced original image from among the pixels constituting the reduced original image Reduced image target block setting means for setting a target block on the reduced original image, which is a target block of the image, and a reduced image search for searching a reduced reference image destination block most similar to the target block on the reduced original image from the reduced reference images And a search range setting means for setting a search range in which the search means searches for the destination block in the reference image based on the position on the reference image corresponding to the position of the destination block on the reduced reference image. More desirable.

  [Operation / Effect] With such a configuration, the destination block R can be searched from the reference image more accurately. According to the above-described configuration, the actual movement of the pattern on the original image is roughly known by using the reduced image. The calculation cost using this reduced image is not so high. After that, if you know where the pattern on the target block on the original image has moved on the reference image, it will be as accurate as if you were searching for the destination block in a wide area of the reference image. Can grasp the destination block.

  Further, in the above-described image processing apparatus, the search unit rotates each of the candidate blocks when searching for a destination block most similar to the target block from candidate blocks that are candidates for the destination block in the reference image. It is more desirable if the similarity is judged about things.

  [Operation / Effect] According to the above-described configuration, even if the image is rotated between the original image and the reference image, the movement destination of the pattern can be accurately calculated.

In the image processing apparatus described above, the target block according to the target setting means may be configured to have a enclave.

In the image processing apparatus described above, the target block according to the target setting means may be configured to exclude some of the peripheral pixels surrounding the pixel of interest.

  [Operation / Effect] With the configuration as described above, the calculation cost of the search means can be reduced. This is because the number of pixels to be calculated is smaller than the search using the target block filled with pixels.

  Further, in the above-described image processing apparatus, the search unit is configured to search for a destination block from each of a plurality of different reference images, and the overlapping unit is configured to search for a destination block searched for in each of the reference images. It is more desirable to generate a fusion block by superimposing each of the overlapping blocks corresponding to each of the target blocks on the original image.

  Further, in the above-described image processing apparatus, the search unit is configured to search for a destination block from each of a plurality of different reference images, and the overlapping unit is configured to search for a destination block searched for in each of the reference images. It is more preferable to generate a plurality of fusion blocks by superimposing each of the blocks to be overlapped corresponding to each of the target blocks on the original image on the original image, and generating a final fusion block by superimposing the plurality of fusion blocks on each other. .

  [Operation / Effect] With the configuration as described above, noise can be reduced based on a plurality of images to be overlapped, so that a higher noise removal effect can be expected.

  Further, in the above-described image processing device, based on the positional relationship between the target block and the movement destination block output by the search unit, by grasping where the target pixel for each of the target blocks has moved on the reference image, When the movement destination on the reference image is recognized for each pixel constituting the original image and the movement direction and movement distance of a certain pixel on the original image are far from the surrounding pixels of the pixel, It is more desirable to have editing means for performing editing that changes the destination block corresponding to the target block so that the destination is the position moved by the same distance in the same direction as the movement of the surrounding pixels.

  [Operation / Effect] With the above-described configuration, it is possible to more accurately identify the block to be overlapped in the image to be overlapped.

  In the above-described image processing apparatus, the search unit is configured to search for a plurality of destination blocks from the reference image, and the overlapping unit corresponds to each of the plurality of destination blocks searched for in the reference image. It is more desirable to generate a plurality of fusion blocks by superimposing each of the blocks to be superposed on the target block on the original image, and generate a final fusion block by superimposing the plurality of fusion blocks on each other.

  [Operation / Effect] With the configuration as described above, noise can be reduced based on a plurality of overlapping blocks, so that a higher noise removal effect can be expected.

  In the above-described image processing apparatus, the search unit corresponds to the target block in the original image when searching for a destination block that is most similar to the target block from candidate blocks that are candidates for the destination block in the reference image. It is desirable that candidate blocks close to the position on the reference image are preferentially recognized as destination blocks.

  [Operation / Effect] With the above-described configuration, it is possible to search for a target block that is more realistic.

  In the above-described image processing apparatus, the image generation unit may be a fusion block having a high degree of variation based on the degree of variation indicating the degree of difference in the pattern reflected in the target block that is the origin of the fusion block and the destination block. It is more desirable to generate a noise-reduced image by overlapping the fusion blocks while adding weights so that the noise-reduced image is reflected lightly.

  [Operation / Effect] With the above-described configuration, the fusion block F that has failed to be superimposed does not strongly affect the noise-reduced image.

  In the above-described image processing apparatus, it is more preferable that the target setting unit operates by distinguishing the pixel on the original image from the pixel that is set as the target pixel and the pixel that is not set.

  [Operation / Effect] According to the above-described configuration, the calculation cost of the generation operation of the fusion block F can be suppressed.

  ADVANTAGE OF THE INVENTION According to this invention, the image processing apparatus which can remove the noise of each flame | frame which comprises a live image reliably can be provided. That is, according to the present invention, a target block is set on the original image, and a search is made for where the target block is reflected in the superimposition target image. According to the present invention, the object image is tracked in units of blocks, and the mistracking is corrected by reflecting a plurality of tracking results for one pixel. , Reliability is significantly higher.

FIG. 3 is a schematic diagram illustrating an overview of image processing according to the first embodiment. FIG. 3 is a schematic diagram illustrating an overview of image processing according to the first embodiment. 1 is a functional block diagram illustrating an overall configuration of an image processing apparatus according to a first embodiment. FIG. 6 is a schematic diagram for explaining the operation of the target setting unit according to the first embodiment. FIG. 6 is a schematic diagram illustrating the operation of a search unit according to the first embodiment. FIG. 6 is a schematic diagram illustrating the operation of a search unit according to the first embodiment. FIG. 6 is a schematic diagram illustrating operations of a search unit and a vector calculation unit according to the first embodiment. 5 is a schematic diagram illustrating a vector map according to Embodiment 1. FIG. FIG. 5 is a schematic diagram for explaining the operation of a block superposition unit according to the first embodiment. FIG. 6 is a schematic diagram for explaining the operation of the image generation unit according to the first embodiment. FIG. 6 is a schematic diagram for explaining the operation of the image generation unit according to the first embodiment. FIG. 6 is a schematic diagram for explaining the operation of the image generation unit according to the first embodiment. It is a schematic diagram explaining operation | movement of the weighting setting part which concerns on Example 1. FIG. It is a schematic diagram explaining operation | movement of the weighting setting part which concerns on Example 1. FIG. It is a schematic diagram explaining operation | movement of the weighting setting part which concerns on Example 1. FIG. It is a schematic diagram explaining operation | movement of the weighting setting part which concerns on Example 1. FIG. It is a schematic diagram explaining operation | movement of the weighting setting part which concerns on Example 1. FIG. It is a schematic diagram explaining operation | movement of the weighting setting part which concerns on Example 1. FIG. FIG. 10 is a schematic diagram for explaining the operation of the vector editing unit according to the first embodiment. FIG. 10 is a schematic diagram for explaining the operation of the vector editing unit according to the first embodiment. FIG. 10 is a schematic diagram for explaining the operation of the vector editing unit according to the first embodiment. FIG. 10 is a schematic diagram for explaining the effect of the vector editing unit according to the first embodiment. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention. It is a mimetic diagram explaining composition of one modification concerning the present invention.

  Hereinafter, each example showing a mode for carrying out the invention will be described. The X-rays in the examples correspond to the radiation of the present invention. The image processing apparatus according to the present invention performs noise reduction processing on an image generated by continuously imaging a subject.

  The configuration of the image processing apparatus 10 of the present invention will be described. The image processing apparatus 10 of the present invention is an apparatus used as an application for reducing noise of a live image. Live image photography is a kind of X-ray photography and aims at photographing a fluoroscopic image as a moving image, and thus a live image becomes a moving image. In such live image capturing, since the subject is exposed to X-rays for a long time, the dose of X-rays used for capturing in order to suppress the exposure dose of the subject can be kept relatively low. ing. Therefore, the live image has a poor S / N ratio and tends to include a lot of statistical noise.

  When a live image V is input to the image processing apparatus 10 of the present invention, a noise-reduced moving image Vα with reduced noise is output from the live image V as shown in FIG. At this time, the image processing apparatus 10 reduces the noise from each frame by performing image processing for each frame constituting the live image V, and generates a noise-reduced video Vα by connecting these in time series. Specifically, the image processing performed on the frame by the image processing apparatus 10 is by canceling noise appearing at random in each frame by superimposing a frame and a frame taken immediately before the frame. Reduce noise on the frame. In this way, noise on the live image V indicated by a plurality of points in FIG. 1 is reduced, and a noise-reduced moving image Vα with improved visibility is generated.

FIG. 2 illustrates an overview of image processing performed by the image processing apparatus 10 according to the first embodiment. It is assumed that the live image V is configured by combining frames that are continuously captured. FIG. 2 shows a state in which a frame with reduced noise is output when the t-th photographed frame is input. In the following description, to be called the t-th frame and the original image I _t, will be the frame obtained by performing a noise reduction process on the t-th frame is called a noise reduced image O _t. Incidentally, the following description of are those when performing noise reduction processing on the original image I _t unless otherwise specified.

The image processing apparatus 10 is premised on a configuration in which noise reduction processing is performed every time a frame is generated. FIG. 2 shows a state in which image processing is performed on the t-th frame. Actually, the image processing apparatus 10 performs noise on the t−1-th frame before performing the operation of FIG. Reduction processing is performed. The t-1th frame is referred to as a reference image It _-1 and a frame obtained by performing image processing on the t-1th frame is referred to as a superimposition target image _Ot-1 . The polymerization object image O _t-1, which at the moment of performing the processing of the original image I _t only intermediate image, when performing image processing on the t-1 th frame, the noise reduction image processing apparatus 10 outputs The image O _t-1 was handled.

As seen by reference to FIG. 2, when the image processing apparatus 10 to the image processing for the t-th frame, the noise reduction by superimposing polymerization object image O _t-1 to the original image I _t which is in the frame An image O _t is generated. However, the image processing apparatus 10, instead of simply overlapping both images, operates to superimpose to determine the mode of superimposition by the reference image I _t-1 comparison with the original image I _t. This mode determination method is the most characteristic part of the present invention, and is handled by the target setting unit 11, the search unit 12, and the vector calculation unit 13 described later. These specific operations will be described later. The target setting unit 11 corresponds to the target setting unit of the present invention, and the search unit 12 corresponds to the searching unit of the present invention.

  FIG. 3 is a functional block diagram for specifically explaining the configuration of the image processing apparatus 10 of the present invention. Hereafter, each part which comprises the image processing apparatus 10 is demonstrated, referring this figure.

<Operation of Target Setting Unit 11>
The original image I _t is input to a target setting unit 11 shown in FIG. Target setting unit 11, as shown in FIG. 4, setting one of the pixels constituting the original image I _t to the pixel of interest. Then, pixels around the target pixel are set as peripheral pixels. Then, a block of pixels composed of the target pixel and peripheral pixels is set as the target block T. In the example of FIG. 4, the target block T is represented by a square range of 5 pixels vertically × 5 pixels horizontally. Among the pixels constituting the target block T, one pixel located at the center is a pixel of interest, and 24 pixels located around the pixel are peripheral pixels.

<Operation of Search Unit 12>
Search unit 12 is configured to find the destination block R that is most similar to the target block T from the different reference time subject was taken bleeds through the image I _t-1 from the original image I _t. The search unit 12 will be described in more detail. Information indicating the target block T, information indicating the position of the target pixel corresponding to the target block T, the original image It _, and the reference image It _-1 are sent to the search unit 12. FIG. 5 shows a state in which the search unit 12 sets the search range r for the target block T on the reference image It _-1 . The search unit 12 sets a rectangular range wider than the target block T around the position corresponding to the target block T on the reference image It _-1 as the search range r. In FIG. 5, the target block corresponding to the target pixel (x, y) is represented by T (x, y), and the search range corresponding to the target block T (x, y) is represented by R (x, y). Yes.

The width of the search range r is determined by the set value held by the search unit 12. The size of the search range r can be the same regardless of the position of the pixel of interest. However, for the pixel of interest located at the end of the original image I _t is, can cause some limits set forth by the set value lies off from the reference image I _t-1, the search range r is narrowed correspondingly It is possible.

Search unit 12 locates from among the target blocks T of the original image I _t (x, y) refers to the pattern similar to the image that crowded appear in the image I _t-1 on the search range r (x, y) . In this way, the search unit 12 performs a pattern search within the limited search range r, so that the search operation is short.

  FIG. 6 specifically shows the pattern matching operation performed by the search unit 12. As shown in FIG. 6, the search unit 12 pays attention to a candidate block C having the same shape and size as the target block T belonging to the search range r, and when the candidate block C and the center of the target block T are overlapped with each other. The absolute value of the subtraction result of the pixel value is calculated for each pixel in the same position in the candidate block C and the target block T, and the sum of the absolute values calculated for each pixel is calculated. This total value is the degree of mutation S (C) indicating how different the blocks are. The degree of variation S (C) indicates that the larger the image reflected in each block is, the larger the degree of mutation S (C) is, and can be expressed as S (C) = Σ | Ti−Ci | as shown in FIG. Ti is a pixel value of each pixel constituting the target block T, and Ci is a pixel value of each pixel constituting the candidate block C. i represents the position of each pixel in the block.

  The search unit 12 calculates the degree of mutation S (C) corresponding to each candidate block C while changing the position of the candidate block C on the search range r (x, y). The right side of FIG. 6 shows a state in which a plurality of mutation degrees are obtained while the candidate block C moves from the upper left to the lower right of the search range r (x, y).

  The search unit 12 selects the candidate block C having the smallest degree of variation, and specifies that the pattern most similar to the image reflected in the target block T is reflected. In this specification, the search unit 12 uses a principle that the degree of variation S (C) decreases as the pattern in the target block T and the pattern in the candidate block C are similar to each other. . If both patterns are exactly the same, the degree of mutation S (C) is 0, and there are no more similar patterns.

  The search unit 12 assumes that the candidate block C most similar to the target block T (x, y) is the destination block R (x, y), and uses the vector calculation unit 13 to display data indicating the position of each block on the image. To send.

<Operation of Vector Calculation Unit 13>
The vector calculation unit 13 calculates a vector v (x, y) indicating the transfer status of the target block T (x, y) based on the search result performed by the search unit 12. The data indicating the position of the target block T (x, y) output from the search unit 12 to the vector calculation unit 13 is specifically the position of the target pixel at the center of the target block T. Specifically, the position of the pixel of interest is coordinates (x, y). The data indicating the position of the destination block R (x, y) output from the search unit 12 to the vector calculation unit 13 is specifically the position of the pixel at the center of the destination block R. This central pixel position is defined as (a, b). The vector calculation unit 13 calculates a vector v (x, y) having a start point (x, y) and an end point (a, b). The vector v (x, y) is has a vector corresponding to the target pixel at the position on the original image I _t (x, y). FIG. 7 shows how the vector calculation unit 13 calculates the vector v (x, y) for the target block (x, y).

<Create vector map mv>
The above description is of the operation of each section 11, 12, 13 for the pixel of interest at the position on the original image _{I t} (x, y). Target pixel is only one of the pixels constituting the original image I _t. In the original image _{I t,} for example, the pixel of interest (x, y) there should be a vector v also corresponding thereto (x + 1, y) for the pixel (x + 1, y) on the right next to the. According to the configuration of the first embodiment has a configuration to calculate the corresponding vector for all the pixels constituting the original image I _t. Therefore, each of the units 11, 12, and 13 repeats the above operation as many times as the number of pixels constituting the original image.

In this manner, each of the pixels constituting the original image I _t, the vector is calculated corresponding thereto. This vector indicates whether the moving far in on the block of the original image I _t is the reference image I _t-1 of the 5 × 5 surrounding the starting point of the vector. Since originally the original image I _t is the reference image I _t-1 is a one obtained by over time to shoot X-rays a subject, a position where crowded-through statue of which are are similar are different from each other. In addition, the images that appear in both images do not simply move in one direction, and may move to the right, not move, or move to the left depending on the portion of the image. The amount of movement is also variable. According to the configuration of the present invention, the complicated movement mode is obtained by calculating a vector indicating the movement mode of the image in units of pixels. This vector indicates the target pixel and the movement destination of the block surrounding it, but may be the destination of the target pixel for convenience.

As understood from the above description, the vector by the same number as the pixels constituting the original image I _t is calculated. For convenience of explanation, these vectors are collectively expressed as a vector map mv as shown in FIG. The vector map mv is arranged following the arrangement of pixels corresponding to each vector. Each vector may be managed by a table or the like instead of the vector map mv.

<Operation of block superposition unit 14>
Vector map mv, the original image _{I t} and polymerization object image _{O t-1} is sent to the block polymerization unit 14. The block superimposing unit 14 is provided for the purpose of reducing noise components reflected in the target block T by superimposing the target block T similar to each target block T. The block polymerization unit 14 corresponds to the polymerization means of the present invention.

Figure 9 shows how a block polymer unit 14 is performing block polymer processing for the pixel at the position on the original image I _t (m, n). The target block T (m, n) corresponding to this pixel coincides with a 5 × 5 range whose center is (m, n). Block polymers portion 14 first, to grasp the target block T (m, n) in the original image _{I t.} Then, the block overlapping unit 14 refers to the vector v (m, n) belonging to the vector map mv and to which position the target block T (m, n) has moved on the overlapping target image O _t−1 . And the superposition target block G (m, n) corresponding to the movement destination is grasped. Then, the target block T (m, n) and the polymerization target block G (m, n) are overlapped to generate a fusion block F (m, n). The center of the polymerization target block G (m, n) is not necessarily located at the position (m, n).

Here, it should be noted that the image used when the vector map mv is calculated is not the same as the image to which the vector map mv is applied. That is, the vector map mv, speaking from the calculation method is not indicate whether the crowded-through in the reference image I _t-1 image of the original image I _t is how moving. Therefore, the vector map mv has nothing to do with the superimposition target image O _t−1 at first glance.

Nevertheless, the vector map mv is used when searching for the superposition target block G from the superposition target image _Ot-1 . Can the block overlapping unit 14 find out a block similar to the target block T by performing such an operation?

The superimposition target image O _t-1 is a noise-reduced image O _t-1 corresponding to the original image It _-1 if the original is corrected. The noise reduced image O _t-1 is generated by superimposing the original image It _-1 and the superimposition target image _Ot-2 . However, the subject images reflected in both images are different. In view of such circumstances, the noise reduced image O _t-1 is generated by fragmenting the image to be overlapped O _t-2 and superimposing the image on the fragment so as to overlap the image of the original image It _-1. Yes. Therefore, the noise reduced image O _t-1 includes the same image as the subject image reflected in the original image It _-1 . The main difference between the two images is whether or not noise is reflected, and the subject images to be reflected are the same.

That is, when performing processing on an original image I _t, if caused to adapt the vector map mv generated based on the reference image I _t-1 to be polymerized image O _t-1, on the original image I _t That is, a block on the polymerization target image O _t-1 similar to the target block T is found.

Then, it seems that the vector map mv should just be produced | generated using superposition | polymerization object image _Ot-1 . Of course there is such a method. However, to know the movement of the target block T, to use the original image I _t equivalent image is desirable. Polymerization object image O _t-1 is an image already noise removal process has been performed, the original image I _t equivalent image is not suitable in that it can not be said the search process. Therefore, in the first embodiment, the vector map mv is generated using the reference image It _-1 .

The above description is of the operation of the block polymer portion 14 of the pixel at the position on the original image I _t (m, n). This pixel is only one of the pixels constituting the original image I _t. In the original image _{I t,} for example it should have a pixel (m, n) fusion blocks corresponding thereto also the pixel (m + 1, n) to the right of the F (m + 1, n). According to the configuration of the first embodiment has a configuration to calculate the corresponding fusion block for all the pixels constituting the original image I _t. Therefore, the block superimposing unit 14 repeats the above operation as many times as the number of pixels constituting the original image.

Note that a predetermined weight may be added when the target block T (m, n) and the superposition target block G (m, n) are overlapped. That is, it is possible to adjust the set value of the weight so that the target block T (m, n) is darker than the superposition target block G (m, n) by two times and is reflected in the fusion block F (m, n). is there. In this case, all the blocks T and G that are repeatedly executed are added with the same weight. Thus, block polymerization unit 14, a polymerization target block G in the same position as the destination block R in the polymerization object image O _t-1 that crowded-through in the same position the object image and the reference image I _t-1 superimposed on the target block T of the original image I _t to produce a fusion block F.

<Operation of Image Generation Unit 15>
The several fusion block F which the block superposition | polymerization part 14 produced | generated is shown. The data indicating the fusion block F is sent to the image generation unit 15 together with the data indicating the position of the fusion block F. The image generation unit 15 corresponds to an image generation unit of the present invention.

Image generating unit 15, the image size by superimposing the fusion block F, as shown in FIG. 10 to generate the same polymerization image St as the original image I _t. A method for overlaying the fusion blocks F at this time will be described. A certain fusion block F (x, y) is composed of 25 pixels arranged in a 5 × 5 rectangle, and the center pixel a is at the position (x, y). The image generation unit 15 arranges the fusion block F (x, y) at the position (x, y) of the superimposed image St. Therefore, the center pixel a is arranged at the position (x, y) of the superimposed image St. Similarly, the image generation unit 15 arranges the fusion block F (x + 1, y) at the position (x + 1, y) of the superimposed image St.

  At this time, the fusion block F (x, y) and the fusion block F (x + 1, y) should overlap each other. The image generation unit 15 performs superposition by adding pixel values for the overlapping portions.

  Then, how many fusion blocks are superimposed on the pixel a at the position (x, y) of the superimposed image St? The fusion block F is composed of 25 pixels. Some fusion blocks F having the pixel a have the pixel a in the lower right corner as shown in FIG. 11, while others have the pixel a at the left of the lower right corner. Some have, others have in the upper left corner. More generally speaking, the pixel a appears in any of 25 types of fusion blocks F (x + m, y + n) expressed when −2 ≦ m ≦ 2 and −2 ≦ n ≦ 2. Since all these fusion blocks F overlap with the superposed image St, 25 fusion blocks F overlap with the pixel a of the superposed image St.

Such polymerization image St is the presence far removed from the original image I _t. That is, the pixel value is higher 25 times the original image I _t. Fusion block F is an image fragment with equivalent pixel values and only the original image I _t it. Since the pixels constituting the superposed image St are formed by stacking 25 fusion blocks F, the pixel value is also about 25 times.

Therefore the image generation unit 15 of the present invention, so that drop to the same level as the level of the original image I _t of pixel values by performing division processing to the polymerization image St as shown in FIG. 12. Note that the number of overlapping fusion blocks F may be less than 25 at the edge of the superimposed image. Therefore, the image generation unit 15 executes the division process of the superposed image St while changing it for each pixel based on the superposition state map indicating how many fusion blocks F are overlapped depending on the place of the superposed image St. However, the image generation unit 15 performs a process of dividing the pixel value by 25 with respect to the majority of the pixels located at the center of the superimposed image St.

The image after the division processing is the noise reduced image O _t . The noise reduced image O _t is a picture such as to remove a noise component from the original image I _t, is output from the image processing apparatus 10 as a frame constituting the noise reduction video V.alpha. Further, the noise reduced image O _t is an image to be overlapped when noise reduction processing is performed on the original image It _{+ 1} . Reference image I _t in the noise reduction processing is the same as the original image I _t according to FIG.

The finished noise reduced image O _t, for this reason will be described since the crowded reflects the same subject image and the original image I _t. Noise reduced image _{O t} is constructed from the original image _{I t} and polymerization object image _{O t-1} are overlapped. Original image visible on captured I _t the subject image polymerized object image O _t-1 rather than in exactly the same image from the subject image bleeds through, partly misalignment has occurred. The image processing apparatus 10 corrects this positional deviation and superimposes both images. That is, when generating a fusion block F, is not corrected in the original image I _t and the position deviation of the polymerization object image O _t-1 have been made. In other words, the polymerization object image O _t-1 is bleeds through the image is corrected are superimposed on the noise-reduced image O _t to correspond to the original image I _t.

On the other hand, the processing of the image processing apparatus 10, Come to and from the original image I _t, is once fine fragments, only it was also returned to single image. Thus, the noise-reduced image O _t is only the noise component of the image visible on captured original image I _t is an image such as removed, the subject image visible on captured both images are the same thing. Such a situation is also the reason that the same subject image is reflected in the superimposition target image O _t-1 and the original image It _-1 .

In this manner, the image generation unit 15, as the target block T is sequentially set while changing the position of the pixel of interest, show-through to the original image I _t by superimposing on the successively image fusion block F A noise-reduced image O _t with reduced noise is generated. In this case, the image generating unit 15 is operated as the position of the fusion block F on the noise reduced image O _t have the same position as the position of the target block T on the original image I _t.

Thus, the image generation unit 15, the original image I by carrying out the different target pixel operation position of the target pixel in the fusion block F to place the fusion block F to be the same as the original image I _t on the image A noise reduced image O _{t in} which noise reflected in _t is reduced is generated. Then, the image generation unit 15 adds the fusion blocks F while overlapping them, and then divides the pixel value of the pixel on the image by the number of integrations indicating how many times the fusion block F is added in the pixel. Thus, the noise reduced image O _t is generated.

<Other configurations>
The above description is the basic operation of the image processing apparatus according to the present invention. The configuration of the present invention can perform image processing for various purposes by performing additional operations in addition to the above-described operations. Therefore, the weighting setting unit 16 and the vector editing unit 17 described below can be operated as appropriate according to necessity.

<Weight setting unit 16>
Weighting setting unit 16 has a configuration related to when the block polymer unit 14 overlapping the polymerization object block G on the target block T and a polymerization object image O _t-1 in the original image I _t. The weight setting unit 16 outputs a set value for changing the superposition mode for each target block T to the block superposition unit 14. The block superposition unit 14 performs superposition in a different manner for each target block T.

  That is, in a configuration in which the weight setting unit 16 is not operated, the block superposition unit 14 superimposes the blocks to be superposed on all the target blocks T by superimposing both blocks T and G based on a constant weight. Operates to overlap at the same density. When the weight setting unit 16 is operated, the block superposition unit 14 can perform the operation of superimposing the superposition target blocks G on a certain target block T, or on another target block T. Can operate to superimpose the superposed blocks G thinner.

  There are a plurality of methods for the operation of the weight setting unit 16. Hereinafter, each method will be described in detail.

<Operation of Weight Setting Unit 16: Method of Changing Weight in Pixel Unit>
FIG. 13 illustrates a case where the weight setting unit 16 is operating in a manner of changing the weight on a pixel basis. In FIG. 13, pixel a of the upper left point of the in the original image _{I t} (x, y) target block T at position (x, y) polymerizing the block G in the pixel A of the upper left point of the (x, y) is It shows how they are superimposed. At this time, the block superposition unit 14 receives the weighting setting value β _A corresponding to the pixel A on the target block T (x, y) from the weighting setting unit 16, and the fusion block F (x, y) The superposition operation is performed so that the pixel at the upper left of the pixel is the sum of the pixel A × (1−β _A ) and the pixel a × β _A. If β _A is 0, the fusion block F is generated without superimposing the pixel a of the superposition target block G (x, y) on the pixel A of the target block T (x, y). Further, if β _A is 0.5, the pixel A of the target block T (x, y) and the pixel a of the overlapping block G (x, y) are overlapped with the same intensity, and the upper left pixel in the fusion block F Is generated. That is, as β increases, the superposition of the pixels a of the block G to be overlapped gradually increases. In the noise-reduced image O _t generated based on the fusion block F generated in this manner, the superposition target image O _t-1 is darkly overlapped or thinly overlapped depending on the part.

  The weighting setting unit 16 refers to a table in which each pixel on the target block T and the setting value β are associated with each other, and sets the setting value β corresponding to a pixel in the target block T that the block overlapping unit 14 is trying to overlap. The block is transmitted to the block superposition unit 14. Therefore, there are as many tables as the number of target blocks T. A common table among the target blocks T is not used.

A pixel A in the original image It belongs to the target block T (x, y). When the target block T (x, y) and the overlap target block G (x, y) are overlapped, the pixel A and the pixel a are overlapped with each other. Weighting setting of this superposition is beta _A. This β _A is determined by the pixel value of the pixel A and the pixel value of the pixel a, as will be described later. Incidentally, the pixel A also belongs to the target block T (x−1, y). When the target block T (x−1, y) and the overlap target block G (x−1, y) are overlapped, a certain pixel on the overlap target image O _t−1 is overlapped on the pixel A. There is no guarantee that the overlapping pixel is the same as the previous pixel a. Since the search unit 12 individually searches for each of the target blocks T, the search result of the target block T (x, y) and the search result of the target block T (x-1, y) are independent of each other. Because.

It is assumed that a pixel α different from the pixel a is superimposed on the pixel A among the pixels on the overlapping block G (x−1, y). Since the pixel a and the pixel α are different from each other, the pixel values are often different from each other. When the target block T (x-1, y) and the polymerization target block G (x-1, y) are overlapped, the pixel A and the pixel α are overlapped with each other. The set value β _α of the superposition weight is determined by the pixel value of the pixel A and the pixel value of the pixel α. That is, even if the same pixel A is overlapped, the set value β may be different if the target block T is different. Therefore, the set value β needs to be calculated for each target block T. The set value β for one pixel A is repeatedly calculated for the number of target blocks T.

FIG. 14 shows a table Tβ (x, y) of setting values β referred to by the weight setting unit 16 when the target block T (x, y) is subjected to the superposition process. In this table Tβ (x, y), set values β _A , β _B ,... Specific to certain pixels A, B. The table Tβ (x, y) is stored in the storage unit 20.

  The weight setting unit 16 generates the table Tβ (x, y) before the block superposition unit 14 performs an operation related to block superposition. A method for generating the table Tβ (x, y) will be described. FIG. 15 shows the relationship between the absolute value | A−a | of the pixel value difference referred to when the weighting setting unit 16 generates the table Tβ (x, y) and the weighting setting value β. Data indicating this relationship is stored in the storage unit 20.

  The absolute value | A−a | is an index indicating how different each pixel is. When this is high, the difference in pixel value is large, and the pixel A is different from the pixel a.

  As can be seen from FIG. 15, the relationship between the absolute value | A−a | and β is set such that β decreases as the absolute value | A−a | decreases. The fact that the absolute value | A−a | is small indicates that the pixel A on the target block T and the pixel a on the overlapping block G are similar to each other. That is, it can be considered that the subject images appearing on both pixels are similar, and even if the pixel a is superimposed on the pixel A so as to be darker, no overlapping of the images occurs. According to the configuration of the present invention, in such a case, β is increased so that the noise on the pixel A is surely eliminated.

On the other hand, the fact that the absolute value | A−a | is large indicates that the pixel A on the target block T and the pixel a on the overlap target block G are different from each other. This means that the subject images appearing on both pixels are different, so that when the operation of superimposing the pixel a on the pixel A is repeated, the subject image is duplicated in the resulting noise-reduced image O _t. It will be reflected. Therefore, according to the configuration of the present invention, in such a case, β is reduced to suppress the overlap of the subject images.

According to the method of changing the weight in units of pixels in this way, the block superposition unit 14 acquires the weight setting value β from the weight setting unit 16, and obtains an individual weight for each pixel constituting the fusion block F. has a structure of polymerizing a pixel of the original image I corresponding to the target block T of the pixels on _t polymerizing object image O _t-1 on the polymerization the block G in association. As the absolute value of the difference between the pixel value of the pixel belonging to the target block T and the pixel value of the corresponding pixel on the overlap target block increases, the overlap weight is set so that the overlap target block G is not succeeded to the fusion block F gradually. Is changed.

According to the above configuration, the weight setting unit 16 uses the pixel value of the pixel belonging to the superimposition target block G on the superimposition target image O _t-1 for setting the setting value β. Instead, the pixel value of the pixel belonging to the movement destination block R on the reference image It _-1 may be used for setting the setting value β.

<Operation of Weight Setting Unit 16: Method of Changing Weight in Block Unit>
The above-described method has a configuration in which the weighting of superposition is changed in units of pixels. The present invention is not limited to such a configuration. FIG. 16 illustrates a case where the weight setting unit 16 is operating in a manner of changing the weight in units of blocks. In FIG 16 shows a state which is brought in the original image _{I t} (x, y) target block T at position (x, y) to be polymerized block G (x, y) is superimposed. At this time, the block superposition unit 14 receives the weighting setting value γ _{x, y} corresponding to the target block T (x, y) from the weighting setting unit 16, and the fusion block F (x, y) is the target. Superposition _| polymerization operation | movement is performed so that it may become the sum of block Tx (1- (gamma) _{x, y} ) and superposition | polymerization object block Gx (gamma) _{x, y} .

If γ _{x, y} is 0, a fusion block F is generated without superimposing the polymerization target block G (x, y) on the target block T (x, y). Further, if γ _{x, y} is 0.5, the target block T (x, y) and the polymerization target block G (x, y) are overlapped with the same intensity to generate the fusion block F. That is, as γ increases, the superposition of the polymerization target blocks G gradually increases.

The weight setting unit 16 refers to a table in which the target block T and the set value γ are related, and transmits the set value γ corresponding to the target block T that the block superimposing unit 14 intends to superimpose to the block superimposing unit 14. It has a configuration. In the noise-reduced image O _t generated based on the fusion block F generated in this manner, the superposition target image O _t-1 is darkly overlapped or thinly overlapped depending on the part.

  FIG. 17 shows a table Tγ of setting values γ referred to by the weighting setting unit 16. In this table Tγ, set values γ (a, b) unique to a certain target block T (a, b) are arranged. The table Tγ is stored in the storage unit 20.

The weight setting unit 16 generates the table Tγ before the block superposition unit 14 performs an operation related to the superposition of blocks. A method for generating the table Tγ will be described. In the original image I _t (x, y) at the position of the target block T (x, y) of the central pixel of a representative pixel M, the polymerization target block G (x, y) and the representative pixel m the center pixel of the To do. FIG. 18 shows the relationship between the absolute value | M−m | of the representative pixel value difference referred to when the weighting setting unit 16 generates the table Tγ and the weighting setting value γ. Data indicating this relationship is stored in the storage unit 20.

The absolute value | M−m | is an index indicating how much the representative pixel of the target block T and the representative pixel of the overlapping block G are different. When this is high, the difference in the representative pixel value is large, and the representative pixel M is different from the representative pixel m. When obtaining the weighting setting value γ (a, b) for the target block T (a, b) with the weighting setting unit 16, the absolute value | M−m | _{a, b} is used. That is, the weight setting unit 16 reads the weight set value γ corresponding to the absolute value | M−m | _{a, b} in the relevance shown in FIG. Set values γ _{a, b} corresponding to a, b).

  As can be seen from FIG. 18, the relationship between the absolute value | M−m | and γ is set such that γ increases as the absolute value | M−m | decreases. The fact that the absolute value | M−m | is small indicates that the target block T and the polymerization target block G are similar to each other. This means that the subject images appearing in both blocks are similar, so that even if the polymerization target block G is superimposed on the target block T so as to be darker, no image overlap occurs. According to the configuration of the present invention, in such a case, γ is increased to ensure that the noise on the target block T is eliminated.

  On the other hand, a large absolute value | M−m | indicates that the target block T and the polymerization target block G are different from each other. This means that the subject images appearing in the two blocks are different, so that if the polymerization target block G is superimposed on the target block T in a deeper manner, the images will overlap. Therefore, according to the configuration of the present invention, in such a case, the overlap of the subject images is suppressed by reducing γ.

The noise-reduced image O _t generated in this way is such that the superimposition target image O _t-1 is darkly superimposed in one part and the noise component is not clearly noticeable, and the superposition target image O _{t in} another part. _-1 is overlapped thinly and the overlap of images is not noticeable. Therefore, according to the configuration including the weighting setting unit 16, it is possible to acquire the noise reduced image O _t with higher visibility.

As described above, according to the method of changing the weight in units of blocks, the block superposition unit 14 acquires the weight setting value γ from the weight setting unit 16 and generates a fusion block F each time the fusion block F is generated. In addition, the target block T on the original image It and the target block G on the target image O _t-1 are superposed. As the absolute value | M−m | of the difference between the representative pixel value on the target block T and the representative pixel value on the superposition target block G increases, the superposition target block is not gradually inherited by the fusion block F. The weight is changed.

According to the above-described configuration, the weight setting unit 16 uses the pixel value of the pixel belonging to the overlap target block G on the overlap target image O _t-1 for setting the set value γ. Instead, the pixel value of the pixel belonging to the movement destination block R on the reference image It _-1 may be used for setting the setting value γ.

  The configuration of the present invention can perform an additional operation in addition to the weight setting unit 16. Hereinafter, the configuration of the vector editing unit 17 related to the generation of the vector map mv will be described. The vector editing unit 17 corresponds to editing means of the present invention.

<Operation of Vector Editing Unit 17>
Vector map mv is a map composed of vector indicating whether the target block T of the original image I _t is moved anywhere on the reference image I _t-1. Since the subject image does not change between the reference image It _-1 and the superimposition target image _Ot-1 , the vector map mv indicates the destination of the target block T on the superposition target image _Ot-1. That is why. However, this vector map mv has the following problems.

  In other words, the vector map mv is generated by selecting a candidate block C that is most similar to the target block T from among a plurality of candidate blocks C and determining that this is the destination block R corresponding to the target block T. Therefore, when the overlapping block G corresponding to the destination block R is superimposed on the target block T, noise on the target block T may be noticeable. Such a phenomenon often occurs particularly when many noise components are distributed on the original image.

  The search unit 12 searches for a candidate similar to the target block T from the candidate blocks C even if it is impossible. If intense noise is reflected in the target block T, the search unit 12 tries to find the noise pattern from the inside of the reference image. Then, the search unit 12 recognizes that the candidate block C most similar to the noise component pattern reflected in the target block T is the destination block R. The vector map mv includes a result obtained by forcibly selecting such a destination block R.

As described above, if the search block 12 is forced to select the destination block R, the destination block R may be selected because the reflected noise component is similar to the noise component of the target block T. Come out. In such a case, when both blocks are overlapped, the noise components are strengthened and the noise becomes conspicuous. Phenomena such noise is constructive, in the original image I _t occur when the S / N ratio is low, in addition to the portion which the noise component in the original image I _t is manifested strong, the subject image to the target block T is It can also occur when it is not reflected.

  The vector editing unit 17 starts from the target pixel of the target block T on the vector map mv, and the center of the destination block R that is meaninglessly similar to the target block T in order to compensate for the disadvantage of the search unit 12. The vector shown in FIG. 6 is searched for and the vector map mv is edited by changing this to a zero vector.

Vector editing unit 17 verifies each vector constituting the vector map mv by using the reference image I _t-1 and the original image I _t. FIG. 19 shows a state in which the vector editing unit 17 verifies the vector at the position (x, y) in the vector map mv. First, the vector editing unit 17 acquires the degree of mutation S (R) _{x, y} between the target block T (x, y) and the destination block R (x, y) from the search unit 12. The upper side of FIG. 19 schematically shows how the degree of mutation S (R) _{x, y} is calculated.

Subsequently, the vector editing unit 17 recognizes a block having the same size and the same shape as the target block T around the position (x, y) on the reference image It _-1 . The block in the reference image I _t-1 in block at the same position as the target block T of the original image I _t, is referred to as a same position block O (x, y). In some cases, the image quality of the noise-reduced image O _t is improved when the fusion block F is generated using the same-position block O rather than the destination block R. This is because when the overlapping block G corresponding to the movement destination block R is superimposed on the target block T, the noise components reflected in each block may be strengthened as described above. In such a case, it is more convenient to generate the fusion block F by superimposing the blocks on the polymerization target image O _{t-1 at} the same position of the target block T because the noise components reflected in each block are eliminated. Good. This is because the noise component reflected in the moving image always fluctuates, and the pattern of the noise component reflected in a specific part of each frame changes over time and does not become the same. The problem is how to determine whether to use the co-located block O to generate the fusion block F. The vector editing unit 17 makes this determination using the mutation degrees S (R) and S (O).

The vector editing unit 17 calculates the degree of mutation S (O) _{x, y} between the target block T (x, y) and the same position block O (x, y). Therefore, S (O) = Σ | Ti−Oi |. Ti is a pixel value of each pixel constituting the target block T, and Oi is a pixel value of each pixel constituting the same-position block O. i represents the position of each pixel in the block. The lower side of FIG. 19 schematically shows how the degree of mutation S (O) _{x, y} is calculated.

The vector editing unit 17 determines the difference between the degree of mutation S (R) _{x, y} related to the destination block R (x, y) and the degree of mutation S (O) _{x, y} related to the same position block O (x, y). Compare with. The same position block O (x, y) is actually a candidate block C when the search unit 12 searches for the destination block R. This co-located block O (x, y) has been selected as the most similar in the end, although it was once a candidate block C that may have been most similar to the target block T (x, y). There is a circumstance that it was not. Therefore, when the same position block O (x, y) and the destination block R (x, y) are at different positions, the degree of mutation S (R) _{x, y} is always greater than the degree of mutation S (O) _{x, y} . Will also be small.

The problem is whether the degree of mutation S (R) _{x, y} is predominantly smaller than the degree of mutation S (O) _{x, y} . The left side of FIG. 20 shows a case where the degree of mutation S (R) _{x, y} is significantly lower than the degree of mutation S (O) _{x, y} . The co-located block O (x, y) is a block that is first considered as a candidate when considering where the target block T (x, y) is on the reference image It _-1 . Since the original image I _t and the reference image I _t-1 is a continuous image, the object image on the original image I _t is the case in the reference image I _t-1 because it should not moved away It is. The fact that the destination block R (x, y) is more similar to the target block T (x, y) than the same-position block O (x, y) means that the destination block R (x, y) The reliability of the selection of y) is high.

On the other hand, as shown on the right side of FIG. 20, when the degree of mutation S (R) _{x, y} is similar to the degree of mutation S (O) _{x, y} , the destination block R (x, y) is the same position block O. It is not similar to the target block T (x, y) over (x, y), and the reliability of selection of the destination block R (x, y) is low.

Specifically, the vector editing unit 17 calculates (x, y) on the vector map mv when the degree of mutation S (O) _{x, y} / variation degree S (R) _{x, y} is equal to or greater than a certain value. If the vector at the position is recognized as having high reliability and the degree of mutation S (O) _{x, y} / variation degree S (R) _{x, y} is equal to or less than a certain value, the vector reliability Is recognized as low.

The vector editing unit 17 performs the above-described verification for all the vectors on the vector map mv, and edits such that the low-reliability vector v (a, b) is changed to a zero vector as shown in FIG. to mv. Therefore, as shown in FIG. 22, the block superposition unit 14 at the position (a, b) where the vector is overwritten, the superposition target block G (x, y) in the superposition target image O _t-1 is the original image I. It becomes the same position as the target block T at _t .

That is, in the image processing device of the present invention, the vector editing unit 17, when the block which are significantly similar to the target block T is not found on the reference image I _t-1, and the target block T of the original image I _t A block on the image to be overlapped O _t-1 corresponding to the same position is overlaid on the target block T. In this way, it is possible to prevent the noise-reduced image O _{t from} being disturbed by superimposing the blocks on the superimposition target image O _t−1 that are meaninglessly similar to the target block T on the target block T.

Block on polymerization object image O _t-1, which corresponds to the target block T the same position on the original image I _t is guaranteed a certain degree of similarity. A reference image I _t-1 to reproduce the subject image bleeds through the original image I _t to be polymerized image O _t-1 is because sequential photographs. In contrast, when searching for a block similar to the target block T from among a plurality of candidate blocks C, the degree of mutation is compared. In this method, the degree of mutation is merely used as an index of similarity, and therefore, there is a high possibility that an error will occur if the block is polymerized only on this basis. Therefore, according to the present invention, when it cannot be said that the degree of mutation S (R) is sufficiently lower than the degree of mutation S (O), it is avoided that the destination block R is similar to the target block T. ing. According to the configuration of the present invention, in such a case, a block on the superimposition target image O _t-1 corresponding to the same position as the target block T is set as a target of superposition, thereby avoiding a large error and reducing the noise reduced image O _t. We are trying to improve the overall reliability.

In this way, the vector editing unit 17 sets the same position block O at the position of the target block T in the reference image It _-1 and the destination block R is superior to the target block compared to the same position block O. If the destination block R found by the search unit 12 is set to the same position block O, the search is performed on the vector map mv that is the output of the search means.

  Each unit 11, 12, 13, 14, 15, 16, and 17 is realized by a CPU that executes various programs. Each unit 11, 12, 13, 14, 15, 16, 17 may be realized by an individual processing device in charge of each.

As described above, according to the present invention, it is possible to provide the image processing apparatus 10 that can reliably remove noise of each frame constituting the live image V. That is, according to the present invention, and set the target block T on the original image I _t, is configured to search whether the target block T is crowded-through where the polymerization object image O _t-1. If so find this way the transfer destination of the target block T from the polymerization object image O _t-1, between images while tracking a subject image on the original image I _t on polymerization object image O _t-1 Since the polymerization can be carried out, the specimen does not appear double due to the polymerization.

  In addition, according to the present invention, the object image is tracked in units of blocks, and a plurality of tracking results are reflected on one pixel to correct mistracking. Compared with this, the reliability becomes much higher.

Further, according to the above-described configuration, after adding the fusion blocks F while overlapping, the pixel value of the pixel on the image is divided by the number of integrations indicating how many times the fusion block F is added in the pixel. Like to do. In this way, a more reliable noise reduction image O _t can be generated.

The reason will be described. Each of the pixel values of pixels constituting the noise reduced image O _t is a result of superposition of the target block T and polymerized target block G. The overlap target block G should have the same subject image as the target block T, but if the search unit 12 misidentifies the destination block R, the overlap target image O _t− polymerization target block is not worthy of G portion of the ₁ occurs when the no small chosen polymerization target block G.

According to the above-described configuration, each pixel constituting the noise reduced image O _t is not configured based on one overlapping block G. That is, each pixel is configured by superimposing different superposition target blocks G in multiple layers. Therefore, even if the recognition target block G is erroneously recognized, the influence on the noise-reduced image O _t is limited. This is because the polymerization target block G related to misrecognition is only one of the multiple polymerization target blocks G that are overlapped.

According to the present invention, after adding the fusion blocks F, the pixel value of each pixel is divided by the number of integrations of the fusion block F to generate a noise reduced image. It becomes substantially the same as the I _t.

The pixel of the polymerization the block G on the polymerization object image O _t-1 to block polymerization unit 14 corresponding to the pixel of the target block T of the original image I _t in a separate weighting for each pixel constituting the fusion block F As the absolute value of the difference between the pixel value of the pixel belonging to the target block T and the pixel value of the corresponding pixel on the overlap target block G increases, the overlap target block is gradually not succeeded to the fusion block F. if superimposed changing weighting of registration as, the motion is severe part of the object image in the original image I _t is, superposition of polymerization object image O _t-1 is to be minimal. In this way, it is possible to more reliably prevent the subject image from being duplicated in the noise reduced image.

  In the method of changing the weight for each pixel, the image processing takes too much time. Therefore, as the absolute value of the difference between the pixel value on the target block T and the pixel value on the overlapping block G increases, the overlapping weight is changed so that the overlapping block G is not succeeded to the fusion block F gradually. In this case, since the weight change is performed in units of blocks, it is possible to perform image processing at a higher speed.

The vector editing unit 17 of the present invention has the following effects. If both the target block T on the original image and the block at the same position on the reference image do not move the subject image reflected in both blocks, or if the subject image does not appear in both blocks The same position block on the image It _-1 is guaranteed to be somewhat similar to the target block T. Therefore, when the fusion block F is generated, the search unit 12 searches for the block corresponding to the target block T from the polymerization target image O _t-1 and superimposes it on the target block T in the polymerization target image O _t−1 . In some cases, the visibility is improved by superimposing corresponding portions as they are.

According to the above-described configuration, in the case of the predetermined condition, the search result by the search unit 12 is discarded, and the portion corresponding to the target block T in the superimposition target image O _t-1 is directly overlapped. The overlapping blocks G that have been certified to have moved, although they are not originally moved because they are similar to each other, are not overlapped to generate the fusion block F, and the visibility of the noise-reduced image is reduced. Will be better.

  The present invention is not limited to the above-described configuration, and can be modified as follows.

According to the configuration of (A) above, had to perform the same search operation for each target block T constituting the original image I _t, the present invention is not limited to this configuration. In other words, the movement destination block R may be searched for a certain target block T, and the search result may be reflected in the subsequent search for the target block T.

FIG. 23 illustrates an operation based on this modification. Target setting unit 11 sets a target block T as a target pixel to pixel p, which is one of pixels constituting the upper original image I _t. The search unit 12 searches for the target block T, for example, using the entire reference image It _-1 as the search range r. By doing in this way, the movement destination on the reference image It _-1 of the target block T can be found reliably. The method for determining the pixel p is not particularly limited. For example, the position of the pixel p may be set in advance or may be selected by the operator. Further, a feature point on the image may be extracted by differential processing and used as a pixel p. A point that stands out on the image, such as a part of the bone of the subject, may be used as the pixel p.

FIG. 24 shows a state where the target block T is set with the pixel q in the vicinity of the pixel p as the target pixel, and the movement destination of this block is being searched from the reference image It _-1 . In this case, the search unit 12 first recognizes whether the pixel q is how far from the pixel p on the original image I _t. At this time, the pixel q is assumed to be separated by kx in the horizontal direction and ky in the vertical direction.

There should be a movement destination of the pixel q around a pixel s that is separated by kx in the horizontal direction and ky in the vertical direction from the movement destination of the point p on the reference image It _-1 . When searching for the target block T having the pixel q as the target pixel from the reference image It _-1 , the search unit 12 does not search from the entire image, but defines the search range r around the pixel s. Perform a search on a range.

That is, the search unit 12 according to the present modification uses the accuracy priority mode for searching for the destination block R for a certain target pixel over a wide range of the reference image It _-1 and the search result of the accuracy priority mode. Based on the two modes of the speed priority mode in which the search for the movement destination block R is performed over a narrow range of the reference image It _{-1 for} the target pixel different from the target pixel that has been processed in the accuracy priority mode. In the speed priority mode, the target pixel that has been processed in the accuracy priority mode and the target pixel found in the accuracy priority mode search on the reference image It _-1 are moved to Based on the positional relationship, the position of the movement destination of the target pixel that is the current search target is predicted on the reference image It _-1 and the movement destination block extends over the range surrounding the predicted position. Operates to search for R.

According to the configuration of the present modification, the search unit 12, but accurate but performed on a limited target pixel on the original image I _t search in accuracy priority mode takes time. The center of the destination block R discovered by this mode should accurately represent the destination of the target pixel. Therefore, say where the certain pixel in the neighborhood of the target pixel on the reference image I _t-1 in the original image I _t, is in the neighborhood of the destination pixel on the reference image I _t-1 of the target pixel Must be. Therefore, when searching for such peripheral pixels, only the periphery of the destination pixel is searched. This is the speed priority mode. With such a configuration, the search unit 12 can perform a search operation having both high speed and accuracy.

(B) According to the above arrangement, the original image I _t and the reference image I _t-1, was the image over time to communicate photographed, the present invention is not limited to this configuration. As shown in FIG. 25, it is possible to use the image _{O t-1} instead of the image _{I t-1} as a reference image. This image O _t-1 is a noise-reduced image for the image It _-1 .

(C) According to the above configuration, the polymerization target image, was the noise reduced image O _t-1 of the image I _t-1, the present invention is not limited to this configuration. As shown in FIG. 26, an image It _-1 can be used instead of the image _Ot-1 as the image to be overlapped. This and the image I _t-1 and the original image I _t, an image over time to photographed communication, towards the image I _t-1 is taken first.

(D) According to the above arrangement, the original image I _t and the reference image I _t-1, an image to be over time to communicate photographed, polymerization target image, the noise reduction for the image I _t-1 Although the image is O _t−1 , the present invention is not limited to this configuration. As shown in FIG. 27, using the image _{O t-1} instead of the image _{I t-1} as a reference image, an image _{I t-1} instead of the image _{O t-1} can be used as a polymerization target image. This image _{O t-1} is the noise-reduced image of the image _{I t-1,} and the image _{I t-1} and the original image _{I t,} an image over time to photographed communication. If the image I _t-1 have been taken earlier than the original image I _t.

That is, the image processing apparatus 10 of the present invention, the (a1) obtained when the image processing on the captured image before the polymerization target image based on the image I _t noise reduced image, (b1) the reference image Can be an image taken before the original image It. The image processing apparatus 10 of the present invention, the (a1) obtained when the image processing on the captured image before the polymerization target image based on the image I _t noise reduced image, (b2) a reference image it can be a corresponding noise-reduced image to the original image I captured images prior to _t.

Similarly, the image processing apparatus 10 of the present invention, (a2) was assumed as the image photographed in front of a polymerization target image based on the image I _t, an image taken in front of the original image I _t and (b1) a reference image can do. The image processing apparatus 10 of the present invention, (a2) was assumed as the image photographed in front of a polymerization target image based on the image I _t, (b2) the reference image, the captured image in front of the original image I _t A corresponding noise-reduced image can be obtained.

  (E) According to the above-described configuration, the weighting is set using the target block T on the original image and the pixel value on the overlap target image, but the present invention is not limited to this configuration. As shown in FIG. 28, the weighting may be set using pixel values of the target block T on the original image and the destination block R on the reference image. As shown in FIG. 29, the same applies to the above-described modified example (D).

  As described above, the image processing apparatus of the present invention can be realized by selecting various modes. Such high selectivity contributes to increasing the degree of freedom of image processing. The image processing apparatus 10 is configured to continuously perform noise reduction processing on frames that constitute a live image. Each noise reduction process may be performed by any of the four methods described with reference to FIGS. 2, 25, 26, 27, 28, and 29.

<Further modifications>
FIG. 30 schematically illustrates image processing according to the first embodiment. Numbers 1 to 9 reflected in the original image indicate images reflected in the image. Although the numbers from 1 to 9 are also reflected in the reference image, the position where the number is reflected is not the same as the position of the original image. The positions where the numbers 1 to 9 above appear in the superimposition target image are the same as the positions on the reference image. In the image processing according to the first embodiment, first, the target setting unit 11 sets a part of the original image to the target block T. In FIG. 30, it is assumed that the target block T is set in a range surrounding the numeral 5. The search unit 12 searches the reference image for the same pattern as the number 5 shown in the target block T. The area found by the search is the movement destination block R. The block superposition unit 14 recognizes a superposition target block G that is a range on the superposition target image corresponding to the destination block R, and superposes this on the target block T on the original image to generate a fusion block F. By executing such an operation at other positions on the original image, a plurality of fusion blocks F are generated. The image generation unit 15 generates a noise reduced image by superimposing these fusion blocks F in multiple layers.

  The above is the outline of the image processing according to the first embodiment. Various modifications in which a part of this configuration is changed will be described.

  (1) In FIG. 30, the movement destination block R and the superposition target block G have the same size, but the present invention is not limited to this configuration. The movement destination block R may be larger than the polymerization target block G. FIG. 31 illustrates the modification. The target setting unit 11 sets a large target block T (the size is the same as the movement destination block R) on the original image.

  The search unit 12 searches the reference image for a range having a pattern reflected in the target block T, and sets the movement destination block R. The block superposition unit 14 recognizes a superposition target block G that is a range on the superposition target image corresponding to the movement destination block R. However, the polymerization target block G at this time is in a smaller range than the movement destination block R. Although the movement destination block R and the superimposition target block G are different in size, they are pixel blocks having the same shape with the center of the block located at the same position between the images. The block overlapping unit 14 sets a small original image overlap block Itg (the size is the same as the overlap target block G) on the original image. The superposed block Itg on the original image is a pixel block having the same shape centered on the target pixel on the original image although it is smaller than the target block T.

  Actually, the superposed block Itg on the original image and the superposed block G are pixel blocks having the same size and the same shape. The block superposition unit 14 generates a fusion block F by superposing the superposition block Itg on the original image and the superposition target block G. As described above, according to this modification, the range of the block G to be superposed on the block superposition unit 14 is narrower than the ranges of the target block T and the destination block R.

  By performing such an operation, it is possible to determine the superposition target block G by evaluating the region outside the superposition target block G, so that the fusion block F can be generated with higher reliability.

  (2) In FIG. 30, the movement destination block R and the polymerization target block G have the same size, but the present invention is not limited to this configuration. The superposition | polymerization object block G can also be set as the structure larger than the movement destination block R. FIG. FIG. 32 illustrates the modification. The target setting unit 11 sets a small target block T (the size is the same as the movement destination block R) on the original image.

  The search unit 12 searches the reference image for a range having a pattern reflected in the target block T, and sets the movement destination block R. The block superposition unit 14 recognizes a superposition target block G that is a range on the superposition target image corresponding to the movement destination block R. However, the polymerization target block G at this time is in a larger range than the movement destination block R. Although the movement destination block R and the superimposition target block G are different in size, they are pixel blocks having the same shape with the center of the block located at the same position between the images. The block overlap unit 14 sets a large original image overlap block Itg (the size is the same as the overlap target block G) on the original image. The superposed block Itg on the original image is a pixel block having the same shape centered on the target pixel on the original image although it is larger than the target block T.

  Actually, the superposed block Itg on the original image and the superposed block G are pixel blocks having the same size and the same shape. The block superposition unit 14 generates a fusion block F by superposing the superposition block Itg on the original image and the superposition target block G. Thus, according to this modification, the range of the block G to be superposed on the block superposition unit 14 is wider than the target block T and the destination block R.

  By performing such an operation, it is possible to generate a noise-reduced image from which noise is further removed. This is because if the fusion block F is made larger, the number of pixels superimposed in a multiple manner when a noise-reduced image is generated increases. FIG. 33 illustrates such a situation. The left side of FIG. 33 shows a case where the fusion block F is small. Since the fusion block F has only a size of 2 × 2 in the vertical direction, a certain pixel on the noise-reduced image has only four fusion blocks F superimposed on each other. On the other hand, the right side of FIG. 33 shows a case where the fusion block F is large. Since the fusion block F has a size of 3 × 3 in the vertical direction, nine fusion blocks F are superimposed on a certain pixel on the noise-reduced image.

  When erasing noise from the original image, it is desirable that the number of fusion blocks F superimposed on the pixels on the noise-reduced image is larger. This is because as the number of fusion blocks F increases, noise on the pixels is averaged and erased. Therefore, the effect of noise removal is higher when the fusion block F is made larger.

  When trying to enlarge the fusion block F, the target block T is matched to the size of the fusion block F according to the idea of the first embodiment. In this case, the fusion block F certainly increases, but the calculation cost of the search unit 12 increases by the amount of the target block T. According to this modification, it is possible to increase the fusion block F without increasing the calculation cost of the search unit 12 and enhance the noise removal effect.

  (3) According to FIG. 5 of the first embodiment, the search range r, which is a range for searching the destination block R, is set based on the position of the target block T. However, the present invention is not limited to this configuration. I can't. The search range r may be determined based on the reduced image. FIG. 34 shows a functional block diagram relating to the determination of the search range r of the present modification. The image reduction unit 9 is configured to reduce the original image and the reference image.

  Hereinafter, the operation of each unit described with reference to FIG. 34 will be described. FIG. 35 shows how the original image and the reference image are reduced by the image reduction unit 9. A reduced original image is referred to as a reduced original image, and a reduced reference image is referred to as a reduced reference image. The original image and the reference image are reduced at the same reduction ratio by the image reduction unit 9 to become reduced images of the same size.

  FIG. 36 shows an operation for setting the search range r on the reference image using the reduced original image and the reduced reference image. That is, in the present modification, the target pixel is set in the reduced original image, and the target block Tc on the reduced original image corresponding to the target pixel is set. At this time, the reduced image target setting unit 11a executes the setting of the target block Tc on the reduced original image. The operation of the reduced image target setting unit 11a is the same as that of the target setting unit 11 of the first embodiment. The target block Tc on the reduced original image is, for example, a square range of 5 pixels vertically × 5 pixels horizontally and is the same as the target block T handled by the target setting unit 11 of the first embodiment. That is, the reduced image target block setting unit 11a performs the same operation as that on the left side of FIG. However, since the target block Tc on the reduced original image in this case is set for the reduced original image, the size of the target block T with respect to the image is relatively larger than in the case of FIG. Data relating to the set target block Tc on the reduced original image is sent to the reduced image search unit 12a.

  The reduced image search unit 12a sets a range including pixels on the reduced reference image corresponding to the target pixel set on the reduced original image as a reduced image search range, and the reduced original image target block Tc within this range. Is searched for a destination block Rc on the reduced reference image having a pattern similar to. The manner in which the reduced reference image upper movement destination block Rc is searched is the same as in FIGS. 5 and 6 of the first embodiment. That is, the reduced image search unit 12a operates in the same manner as in FIGS. 5 and 6 of the first embodiment. In the operation of the reduced image search unit 12a, it is not always necessary to set a range (search range) for searching the reduced reference image upper block Rc. This is because the search for the reduced reference image upper block Rc in this case is performed with respect to the reduced reference image having a small image size, and the calculation cost for the search is not so high.

  Data relating to the reduced reference image upper block Rc is sent to the search range setting unit 9. The search range setting unit 9 sets a search range r on the reference image before reduction based on the reduced reference image upper movement destination block Rc found on the reduced reference image. Therefore, the search unit 12 re-partitions the range once searched by the reduced image search unit 12a and searches for the target block T again. In this case, the search unit 12 operates using the target block T having a resolution higher than that of the target block Tc on the reduced original image, so that the block search can be performed more strictly.

  The operation of the search range setting unit 9 will be described. The range corresponding to the movement destination block Rc on the reduced reference image should also be on the reference image before the reduction. The range at this time should be a range in which the destination block R on the reduced reference image is enlarged. The search range setting unit 9 sets the search range r based on the position on the reference image corresponding to the position of the reduced image upper movement destination block Rc. At this time, the search range r is set wider than the range (corresponding range) on the reference image corresponding to the reduced reference image upper block Rc. This is because when the search unit 12 searches for the destination of the target block T on the reference image, it is necessary to prepare for the case where the destination is outside the corresponding range.

  The reduced image target setting unit 11a, the reduced image search unit 12a, and the search range setting unit 9 reduce the entire area of the reduced original image while setting each of the pixels on the reduced original image as target pixels in order. A search range r on the previous reference image is set. Data associated with each pixel of the reduced original image and the search range r is sent to the search unit 12.

  Each unit 8, 9, 11a, 12a is realized by a CPU that executes various programs. Each unit 8, 9, 11a, 12a may be realized by an individual processing device in charge of each.

  FIG. 37 shows operations of the target setting unit 11 and the search unit 12 in this modification. First, as described with reference to FIG. 5, the target setting unit 11 sets a target block T corresponding to a certain target pixel on the original image. The target block T has, for example, a square range of 5 pixels vertically × 5 pixels horizontally. The search unit 12 receives data relating to the target block T from the target setting unit 11 and calculates where the target pixel corresponding to the target block T is located in the reduced original image. Since a search range r is set for each pixel of the reduced image, the search unit 12 determines which pixel on the reduced image the target pixel corresponding to the target block T set by the target setting unit 11 corresponds to. For example, the search range r corresponding to the target block T set by the target setting unit 11 can be known. FIG. 37 shows a state in which the search unit 12 determines the search range r corresponding to the target block T on the reference image. The portion outside the search range r in the reference image is a search exclusion region, and the search unit 12 ignores this region and executes the search operation for the destination block R. This is the same as described with reference to FIG. The search unit 12 searches for a destination block R from the search range r. The pixel block of the overlap target image at the same position of the movement destination block R on the reference image becomes the overlap target block G.

  That is, according to the present modification, the image reduction unit 9 that generates the reduced original image and the reduced reference image by reducing the original image and the reference image, and the target pixel and the reduced original image among the pixels constituting the reduced original image A reduced image target setting unit 11a that sets a reduced original image target block Tc that is the upper target block, and a reduced reference image upper movement destination block Rc that is most similar to the reduced original image target block Tc among the reduced reference images. A search range in which a reduced image search unit 12a to search for and a range on the reference image corresponding to the reduced reference image upper destination block Rc are set as a search range r that is a range in which the search unit 12 searches for the destination block R in the reference image A setting unit 9 is provided. The image reduction unit corresponds to the image reduction unit of the present invention and corresponds to the reduced image target block setting unit of the present invention. The reduced image search unit corresponds to the reduced image search unit of the present invention, and the search range setting unit corresponds to the search range setting unit of the present invention.

  With the configuration as in this modification, the destination block R can be searched from the reference image more accurately. According to FIG. 5 of the first embodiment, the search range r is determined by expanding the range on the reference block corresponding to the target block T on the original image. Such an operation is based on the expectation that the pattern on the target block T on the original image must be near the position on the reference image corresponding to the target block T. This expectation can be disappointing. On the other hand, according to the configuration of the present modification, the configuration is such that the movement of the pattern on the original image is roughly known by actually using the reduced image. The calculation cost using this reduced image is not so high. In this modification, it is grasped where the pattern on the target block T on the original image has moved on the reference image thereafter. Therefore, it is possible to grasp the destination block R with accuracy as if the destination block R is searched in a wide area of the reference image.

  In the description of FIG. 36, the search range r is determined directly from one reduced image set (reduced original image and reduced reference image). However, the present modification is not limited to this configuration. It is also possible to prepare reduced original image sets with different scales and determine the search range r in multiple steps. In the case of this configuration, the search range r on the reference image is determined by repeating the operations of FIGS. 36 and 37 in order from the smallest reduced image set. The search range of the target block Tc on the reduced original image for the reduced image set reduced to the second smallest size is determined based on the reduced image set reduced to the smallest size. This operation is repeated while changing the scale image set, and finally the search range r is determined for the image set (original image and reference image) before the scale. The size of the target block Tc on the reduced original image can be made the same size as the target block T. In this way, the destination block R can be grasped with accuracy as if the destination block R is being searched in a wider area of the reference image.

  (4) In the method of the first embodiment, it is assumed that the pattern of the target block T on the original image moves without rotating on the reference image, but the pattern of the target block T rotates on the reference image. It is also possible to adopt a configuration that takes into account the possibility of being reflected. In this configuration, a plurality of target blocks T are prepared for one target pixel. FIG. 38 illustrates the operation of the target setting unit 11 according to this modification. As shown in FIG. 38, first, the search unit 12 sets a candidate block C that is a candidate for the destination block R as described in FIG. The candidate block C set at this time is called a basic candidate block, and the operation at this time is called a basic candidate block setting process.

  The search unit 12 rotates a pattern reflected in the basic candidate block around the target pixel by a predetermined rotation amount to generate a new candidate block. The block generated at this time is called a rotation candidate block Cr. The example of FIG. 38 shows a state in which seven types of rotation candidate blocks Cr are generated by repeating the operation of rotating a certain basic candidate block by 45 ° clockwise. For example, the rotation amount of the basic candidate block may be set to 15 °, for example, so that 23 types of rotation candidate blocks Cr may be generated, or one rotation amount may be reduced to generate more rotation candidate blocks Cr. You may do it.

  FIG. 39 shows a state in which the search unit 12 executes search processing based on eight types of candidate blocks (one basic block and seven rotation blocks). The operation at this time is the same as the operation described in FIG. The difference from FIG. 6 is that the degree of mutation S (C) for eight candidate blocks C is calculated for one target block T. That is, according to the present modification, in response to the increase of the candidate block C by 8 times, the operation for calculating the mutation degree S (C) is increased by 8 times. The search unit 12 obtains the degree of mutation S (C) for each combination of the candidate block C and the target block T, searches for the combination having the lowest degree of mutation S (C), and sets this as the destination block R. In the destination block R, a rotation of the pattern on the target block T is reflected. The pixel block of the overlap target image at the same position of the movement destination block R on the reference image becomes the overlap target block G. The rotation of the pattern on the target block T is also reflected in the polymerization target block G.

  If the polymerization target block G is directly superimposed on the target block T as it is, the rotated patterns are reflected twice. In consideration of this point, the block superposition unit 14 performs a rotation process on the polymerization target block G, and superimposes the processed polymerization target block G on the target block T. The rotation amount of the block superimposing unit 14 at this time can be known by the search unit 12 which of the eight types of candidate blocks is the destination block R. In the example of FIG. 39, since the rotation candidate block Cr obtained by rotating the basic candidate block by 45 ° counterclockwise is recognized as the movement destination block R, the block superposition unit 14 sets the superposition target block G to 45 counterclockwise. When rotated, the direction of the polymerization target block G can be matched with the direction of the target block T.

  When the search unit 12 of this modification finds a destination block R that is most similar to the target block T from among the candidate blocks C in the reference image, the similarity determination is also performed for the rotated candidate blocks C. do.

  According to this modification, even if there is an image rotation between the original image and the reference image, the movement destination of the pattern can be calculated accurately.

(5) According to the first embodiment, the search unit 12 calculates the degree of mutation S (C) as shown in FIG. 6, but the present invention is not limited to this configuration. The degree of mutation S (C) may be calculated by the following equations.
S = (α · Σ (| Ti−Ci |) ^β ) ^γ where α, β and γ are arbitrary coefficients S = (α · Σ (Ti−Ci) ^2β ) ^γ where α and γ are arbitrary coefficients 2β is an even number S = (α · (| ΣTi−ΣCi |) ^β ) ^γ where α, β and γ are arbitrary coefficients S = (α · (ΣTi−ΣCi) ^2β ) ^γ where α and γ are arbitrary The coefficient 2β is an even number. It is also possible to calculate the degree of mutation S (C) using a normalized cross-correlation function, mutual information, conditional entropy, joint entropy, and the like.

(6) In FIG. 2, FIG. 25, FIG. 26, FIG. 27, FIG. 28, FIG. 29, the reference image is It _-1 or _Ot-1 , and the image to be overlapped is It _-1 or _Ot-1. However, the present invention is not limited to this configuration. Reference image may be an _{I t-2, I t-} 3 image input in the past than _{I t-1 as, O t-2, O t} -3 O t-1 past than as It is good also as the image output to. This situation is the same for the polymerization target image.

  (7) According to the first embodiment, the shape of the target block T is a square as shown in the upper part of FIG. 40, but the present invention is not limited to this configuration. The shape of the target block T can be a rectangular shape as shown in the middle of FIG. Further, the shape of the target block T may be a shape obtained by rotating the square target block T shown in the first embodiment.

  Furthermore, although the target block T of the first embodiment has a configuration without an enclave, the target block T can be configured to have an enclave as shown in the lower part of FIG. That is, a pixel group configured by arranging elongated regions extending in the horizontal direction in the vertical direction can be used in place of the target block T of the first embodiment, and the elongated regions extending in the vertical direction can be used in the horizontal direction. A pixel group arranged in the direction can be used in place of the target block T of the first embodiment. Further, a pixel group configured by arranging pixels in a checkered pattern can be used in place of the target block T of the first embodiment.

  If the search operation is performed using a pixel group having an enclave as shown in the lower part of FIG. 40, the calculation cost of the search unit 12 can be reduced. This is because the number of pixels to be calculated is smaller than the search using the target block T filled with pixels. In this case, the fusion block F has enclaves. However, as the target blocks T are set one after another, the fusion blocks F are added one after the other while overlapping each other on the image. A noise reduction processed image in which is eliminated is generated.

  Similarly, the target block T can be configured by excluding a part of peripheral pixels surrounding the target pixel. Even in this case, the fusion block F has excluded pixels. However, as the target block T is set one after another, the fusion block F is added one after the other while overlapping on the image. Thus, a noise reduction processed image in which excluded pixels are eliminated is generated. Even with such a configuration, the calculation cost of the search unit 12 can be reduced.

  (8) In the configuration of the first embodiment, one reference image and one superimposition target image are prepared for each original image, but the present invention is not limited to this configuration. A plurality of reference images and a plurality of overlapping images may be prepared for each original image. FIG. 42 shows an example of such a configuration. In the configuration of FIG. 42, the setting of the superposition target block G for the target block T on the original image is performed for different superposition target images, and a single fusion block F is generated from the target block T and a plurality of superposition target blocks G.

42, first, between the original image, the reference image 1 (for example, the image I _t-1 ), and the image to be overlapped 1 (for example, the image O _t-1 ), the overlap of the block G to be overlapped in the same manner as described in FIG. I am trying to set it. The polymerization target block G at this time is referred to as a polymerization target block G1. 42, between the original image, the reference image 2 (for example, the image I _t-2 ), and the image to be overlapped 2 (for example, the image O _t-2 ), the overlap of the block G to be overlapped is the same as the description in FIG. I am trying to set it. The polymerization target block G at this time is referred to as a polymerization target block G2. The two overlap target blocks G are set for the same target pixel on the original image.

  The target block T, the polymerization target block G1, and the polymerization target block G2 are sent to the block polymerization unit 14. The block superposition unit 14 averages and superimposes the sent blocks to generate a single fusion block F.

  A median filter can also be used as a method for generating the fusion block F performed by the block superposition unit 14 at this time. That is, when determining the pixel value of the pixel at a certain position of the fusion block F, the block overlapping unit 14 extracts and compares the pixels at the same position of the target block T, the overlapping target block G1, and the overlapping target block G2. The block superposition unit 14 recognizes a pixel having an intermediate value among the pixels derived from the target block T, the superposition target block G1, and the superposition target block G2, and uses the value as the pixel value of the pixel at that position of the fusion block F. And The block superposition unit 14 completes the fusion block F by performing this operation over the entire area of the fusion block F. The noise component that appears in the image often has an extreme pixel value. The purpose of using the median filter is to prevent noise components from being inherited by the fusion block F.

  As described above, the search unit 12 of the present modification is configured to search for the destination blocks R1 and R2 from among a plurality of different reference images, and the block superposition unit 14 searches for each of the reference images. The fusion block F is generated by superimposing each of the superposition target blocks G1, G2 corresponding to each of the moved destination blocks R1, R2 on the target block T on the original image.

  With such a configuration, noise can be reduced based on a plurality of images to be overlapped, so that a higher noise removal effect can be expected.

  In the case of FIG. 42, two sets of reference images and superimposition target images are prepared. However, the present modification is not limited to this configuration. Similar processing may be performed between three or more sets. In this case, the fusion block F is generated by superposing three or more polymerization target blocks G1, G2, G3,.

  (9) In the configuration of the first embodiment, one reference image and one superimposition target image are prepared for each original image, but the present invention is not limited to this configuration. A plurality of reference images and a plurality of overlapping images may be prepared for each original image. FIG. 43 shows an example of such a configuration. In the configuration of FIG. 43, the setting of the superposition target block G for the target block T on the original image is performed for different superposition target images, and the fusion block F is generated from each of the target block T and the superposition target block G. And according to this modification, the produced | generated several fusion block F is further superposed | polymerized and the single fusion block F is produced | generated. The fusion block F generated in this way is the final fusion block F for a certain target pixel on the original image, and is the basis of the noise-reduced image.

In FIG. 43, first, the fusion block F is generated between the original image, the reference image 1 (for example, the image I _t-1 ), and the superimposition target image 1 (for example, the image O _t-1 ) as described in FIG. Like to do. The fusion block F at this time is referred to as a fusion block F1. In FIG. 43, the fusion block F is generated between the original image, the reference image 2 (for example, the image I _t-2 ), and the superimposition target image 2 (for example, the image O _t-2 ) in the same manner as described with reference to FIG. Like to do. The fusion block F at this time is called a fusion block F2. The generation of the two fusion blocks F is performed for the same target pixel on the original image.

  Each of the fusion blocks F is sent to the block superposition unit 14. The block superposition unit 14 averages and superimposes the sent fusion blocks F to generate a single fusion block F.

  A median filter can also be used as a method for generating the fusion block F performed by the block superposition unit 14 at this time. That is, when determining the pixel value of the pixel at each position of the fusion block F, the block superimposing unit 14 extracts and compares the pixels at the same position of the fusion block F. The block superposition unit 14 recognizes a pixel having an intermediate value among the pixels derived from each of the fusion blocks F, and sets the value as the pixel value of the pixel at the position of the final fusion block F. The block superposition unit 14 performs this operation for the entire region of the fusion block F, and completes the final fusion block F. However, in the case of FIG. 43, since there are only two fusion blocks F to be overlapped, the operation using the median filter cannot be performed. In order to use the median filter, it is necessary to generate three or more fusion blocks F using three or more types of reference images and their corresponding images to be superposed. The noise component that appears in the image often has an extreme pixel value. The purpose of using the median filter is to prevent noise components from being inherited by the fusion block F.

  As described above, the search unit 12 of the present modification is configured to search for the destination blocks R1 and R2 from among a plurality of different reference images, and the block superposition unit 14 searches for each of the reference images. A plurality of fusion blocks F1, F2 are generated by superimposing the target block T on the original image on each of the superposition target blocks G1, G2 corresponding to each of the destination blocks R1, R2 thus generated. The final fusion block F is generated by superimposing F2 on each other.

  With such a configuration, noise can be reduced based on a plurality of images to be overlapped, so that a higher noise removal effect can be expected.

  In the case of FIG. 43, two sets of reference images and superimposition target images are prepared. However, the present modification is not limited to this configuration. Similar processing may be performed between three or more sets. In this case, the final fusion block F is generated by superposing three or more fusion blocks F1, F2, F3.

  (10) The vector editing unit 17 according to the first embodiment determines whether the subject image reflected in both blocks in the target block T on the original image and the candidate block on the reference image corresponding to the target block T does not move or When the subject image does not appear in both blocks in the first place, the movement destination block R that is most similar to the target block T is selected, so that it is recognized that there is movement without originally moving. As a countermeasure for this, the configuration is such that the vector on the vector map mv is converted to a zero vector. However, instead of or in addition to this function, a configuration for correcting variation in the vector can also be added.

  The inside image keeps moving while shooting a movie. The moving direction of the image should be aligned in almost the same direction when viewed in pixel units. The reason why it is reflected in the video is that small particles do not move freely, but are perspective images of large structures such as the subject. Although the direction of movement of the entire original image is not the same, if the original image is viewed microscopically, the movement of the image between adjacent pixels should be almost the same.

  FIG. 44 illustrates the vector map mv generated by the vector calculation unit 13. The vector map is obtained by mapping a vector indicating where each pixel constituting the original image has moved on the reference image. When a part of the vector map mv is taken out, similar vectors are located in adjacent pixels. A closer observation of FIG. 44 reveals that pixels having vectors that are similar to each other are interleaved with pixels having vectors that differ in only one direction. Such a vector should be oriented in the same direction as an adjacent pixel. However, the vector calculation unit 13 cannot correctly calculate the vector direction due to the influence of noise components included in the original image and the reference image. Such a vector is called a defective vector.

  According to this modification, this defective vector can be erased by correction. As shown in FIG. 45, the vector editing unit 17 sets a processing range in a part of the vector map mv. This processing range is set so narrow that the vectors within this range can be expected to be approximately the same. The vector editing unit 17 calculates an average vector v (ave) by averaging the vectors in the pixels within the processing range. Then, the vector editing unit 17 compares the average vector v (ave) with each of the pixel vectors in the processing range to find a defective vector from the processing range. This operation is called vector variation analysis.

Details of the vector variation analysis will be described. The X component of the average vector v (ave) is X _ave and the Y component is Y _ave . A vector within the processing range is a vector v (i), and i takes a value from 1 to N. Therefore, N indicates the number of vectors in the processing range. Let the X component of the vector v (i) be X _i and the Y component be Y _i . Based on these, the vector editing unit 17 obtains a vector variation allowable amount for the X component and the Y component. The variation tolerances KX and KY can be calculated by the following formulas.

  The vector editing unit 17 determines whether the X component of the vector v (i) does not fit in the variation allowable amount KX or the Y component of the vector v (i) within the processing range does not fit in the variation allowable amount KY. In this case, the vector v (i) is recognized as a defective vector. Then, the vector editing unit 17 replaces the defective vector with the average vector v (ave).

That is, the vector editing unit 17 of the present modification example determines where the target pixel of each target block T has moved on the reference image based on the positional relationship between the target block T and the destination block R output by the search unit 12. To recognize the movement destination on the reference image for each pixel constituting the original image, so that the movement direction and movement distance of a certain pixel on the original image are far from the surrounding pixels of the pixel. If it is, the destination block R corresponding to the target block T is changed by editing the vector map mv so that the destination of the pixel moves to the position moved by the same distance in the same direction as the movement of the surrounding pixels. Edit.

  With such a configuration, the influence of the noise component superimposed on the original image and the reference image that appeared in the vector map mv is mitigated, so that the overlapping block G in the overlapping image can be more accurately identified. .

  (11) According to the configuration of the first embodiment, the single destination block R is recognized from the reference image as shown in FIG. 30, but the present invention is not limited to this configuration. A plurality of destination blocks R may be recognized from the reference image.

  FIG. 46 illustrates the configuration of this modification. The search unit 12 searches for where on the reference image the target block T set by the target setting unit 11 has moved. At this time, the search unit 12 sets a block that is most similar to the target block T among the plurality of candidate blocks as the movement destination block RA. Then, the search unit 12 sets the second most similar candidate block as the movement destination block RB, and sets the third most similar candidate block as the movement destination block RC. Then, the fourth similar candidate block is set as a movement destination block RD.

  In this way, in the method shown in FIG. 30, the search unit 12 is configured to select only one destination block R from the reference image. However, in this modification, the search unit 12 The configuration is such that four destination blocks R are selected from the inside. The block superposition unit 14 selects superposition target blocks GA, GB, GC, GD corresponding to the destination blocks RA, RB, RC, RD from the superposition target images, and the superposition target blocks GA, GB, GC, GD are selected. A plurality of fusion blocks FA, FB, FC, and FD are generated by superposing the target block T of the original image on each. The block superposition unit 14 then superimposes the fusion blocks FA, FB, FC, and FD to generate a final fusion block F.

  When superposing the fusion blocks FA, FB, FC, and FD, the block superposition unit 14 has a mutation degree S (of the destination blocks RA, RB, RC, and RD that are the basis of the fusion blocks FA, FB, FC, and FD. R). This mutation degree S (R) is calculated when the destination blocks RA, RB, RC, and RD are still candidate blocks (see FIG. 6), and the mutation of the destination blocks RA, RB, RC, and RD If the degrees S (R) are SA, SB, SC, and SD, respectively, there is a relationship of SA <SB <SC <SD.

Based on the degree of mutation SA, SB, SC, SD, the block superposition unit 14 superposes the fusion blocks FA, FB, FC, FD by adding weights so that the closer the similarity, the deeper the fusion block F appears. To do. For example, the final fusion block F _final can be generated based on the following mathematical formula.
F _final = (SA- ¹ · FA + SB- ¹ · FB + SC- ¹ · FC + SD- ¹ · FD) / (SA- ¹ + SB- ¹ + SC- ¹ + SD- ¹ )
However, FA, FB, FC, and FD represent the fusion blocks FA, FB, FC, and FD, and SA, SB, SC, and SD represent the degree of mutation corresponding to the fusion blocks FA, FB, FC, and FD. ing. (SA ⁻¹ + SB ⁻¹ + SC ⁻¹ + SD ⁻¹ ) is a constant for normalization.

  That is, the search unit 12 of this modification is configured to search for a plurality of destination blocks from the reference image, and the block superposition unit 14 includes a plurality of destination blocks RA, RB, A plurality of fusion blocks FA, FB, FC, and FD are generated by superimposing the target block T on the original image on each of the polymerization target blocks GA, GB, GC, and GD corresponding to each of RC and RD. The final fusion block F is generated by superimposing the blocks FA, FB, FC, and FD on each other.

  With such a configuration, noise can be reduced based on a plurality of overlapping blocks G, so that a higher noise removal effect can be expected.

  In FIG. 46, four destination blocks RA, RB, RC, and RD are set. However, the destination block to be set can be increased or decreased. In this case, the number of blocks to be overlapped also increases / decreases as the movement destination block increases / decreases.

(12) According to FIG. 6, the degree of mutation S (C) is calculated by the pixels constituting the target block T and the candidate block C, but the present invention is not limited to this configuration. The degree of mutation S (C) may be calculated in consideration of the displacement of both blocks. If the position of the target block T on the original image and the position of the candidate block C on the reference image are too shifted, it is unlikely that the two blocks are similar. This is because, when shooting a moving image, it is considered that the positions where the subject image appears in each frame are similar. According to the present modification, the degree of mutation S (C) can be calculated taking such circumstances into consideration.
The search unit 12 in this modification can calculate the mutation degree S (C) based on the following mathematical formula.

  Here, ω is a constant, and v (t) is a vector of pixels on the vector map corresponding to the target pixel on the original image.

  That is, when the search unit 12 of this modification finds a destination block R most similar to the target block T from candidate blocks C that are candidates for the destination block R in the reference image, the target block T in the original image The candidate block C close to the position on the reference image corresponding to is preferentially recognized as the destination block R.

  In this way, the candidate block C whose position on the reference image is close to the position of the target block T on the original image is preferentially recognized as the destination block R. According to this modification, it is possible to search for the target block T that is more realistic.

  (13) In the configuration of the first embodiment, the fusion blocks F are equally overlapped to generate a noise-removed image, but the present invention is not limited to this configuration. The noise removal image may be generated by superimposing the fusion block F while weighting. In order to describe this modification, first, the configuration of the first embodiment will be described. First, it is assumed that there are an original image, a reference image, and a superimposition target image as shown in FIG. The number “5” appears in the center of the original image, and the number “5” appears in the reference image at a position shifted from the center. In the superimposition target image, the numeral “5” is reflected at the same position as the reference image.

  According to the configuration of the first embodiment, as shown in FIG. 48, the target block T is set one after another while changing the target pixel, and the corresponding fusion block F is calculated. Each fusion block F is overlapped with the same weight to generate a noise reduced image.

  According to this modified example, as shown in FIG. 49, the operation is the same as that of the first embodiment until the fusion block F is generated one after another while changing the target pixel. When comparing the fusion blocks F generated at this time, it is noticed that there is a fusion block F in which the image is not superposed well. Depending on the target block T on which the fusion block F is based, the “5” image on the reference image may not overlap well, or noise may not be sufficiently reduced in the fusion block F. In view of this, the present modified example has a configuration in which the blocks are multiplexed and overlapped while thinning out the fusion blocks F that have failed to be generated so that the noise-reduced image is not affected by the failure.

  When superimposing the fusion blocks F, the block superposition unit 14 refers to the degree of mutation S (R) of the destination block R that is the basis of each fusion block F. The degree of mutation S (R) is calculated when the destination block R is still a candidate block (see FIG. 6), and tends to take a higher value as the failure degree of the fusion block F increases.

Based on the degree of mutation S (R), the block superposition unit 14 weights the fusion block F so that the higher the success degree, the deeper the fusion block F appears, and the multiple superposition is performed. That is, 25 fusion blocks F1 to F25 are superimposed on the pixels constituting the noise reduced image. The pixel value a of this pixel is obtained as follows, for example.
a = (S1 ⁻¹ · F1 + S2 ⁻¹ · F2 +... + S25 ⁻¹ · F25) / (S1 ⁻¹ + S2 ⁻¹ +... + S25 ⁻¹ )
However, S1, S2,..., S25 represent the degree of mutation corresponding to the fusion blocks F1 to F25. (S1 ⁻¹ + S2 ⁻¹ +... + S25 ⁻¹ ) is a constant for normalization.

  That is, the image generation unit 15 according to the present invention has a degree of variation S (R indicating the degree of pattern difference between the target blocks T1 to T25 that are the origins of the fusion blocks F1 to F25 and the destination blocks R1 to R25. ) Based on 1 to S (R) 25, the fusion block F having a higher degree of mutation S (R) is overlapped with the fusion blocks F1 to F25 while being weighted so as to be reflected in the noise-reduced image. A noise-reduced image is generated.

  With such a configuration, the fusion block F that has failed to be superimposed does not strongly affect the noise-reduced image. In addition, unlike the configuration in which the fusion block F that has failed to be superimposed is completely thinned out, the fusion block F that has failed to be superimposed also appears thinly in the noise-reduced image. The number of fusion blocks F stacked in multiple layers is not reduced, and the noise reduction capability can be prevented from being lost as much as possible.

  (14) In the configuration of the first embodiment, the target block T is set for all the pixels constituting the original image. However, the present invention is not limited to this configuration. As shown in FIG. 50, the target block T can be set only for a part of the pixels constituting the original image (pixels indicated by hatching in FIG. 50). That is, according to the present modification, the target setting unit 11 operates by distinguishing the pixel on the original image from the pixel that is set as the target pixel and the pixel that is not set. In the case of FIG. 50, since the pixel designated as the target pixel is arranged in a checkered pattern, the setting of the target pixel is half. By adopting such a configuration, the calculation cost of the operation of generating the fusion block F can be reduced to half.

  (15) The present invention may be configured to include a plurality of the above-described modifications.

  As described above, the present invention is suitable for medical devices.

11 Target setting section (target setting means)
12 Search unit (search means)
14 Block polymerization part (polymerization means)
15 Image generation unit (image generation means)
17 Vector editor (editing means)

Claims (23)

In an image processing apparatus that performs noise reduction processing on an image generated by continuously imaging a subject,
Target setting means for setting a target block composed of a pixel of interest and peripheral pixels surrounding the pixel of interest from among pixels constituting the original image in which the subject image is reflected;
Search means for searching for a destination block most similar to the target block from a reference image in which a subject photographed at a time different from the original image is shown;
Superimposing means for generating a fusion block by superimposing a polymerization target block at the same position as the movement destination block on the target block on the original image in a polymerization target image in which the subject image is reflected at the same position as the reference image When,
An image that generates a noise-reduced image in which noise reflected in the original image is reduced by superimposing the fusion blocks on the image one after another as the target block is set one after another while changing the position of the target pixel. Generating means,
The image processing device operates such that the position of the fusion block on the noise-reduced image is the same as the position of the target block on the original image. The image processing apparatus according to claim 1.
The image generation means adds the fusion blocks while overlapping, and then divides the pixel value of the pixel on the image by an integration count indicating how many times the fusion block is added in the pixel. An image processing apparatus that generates the noise-reduced image. The image processing apparatus according to claim 1 or 2,
The superimposing means superimposes the pixel of the target block on the superimposition image corresponding to the pixel of the target block on the original image with individual weighting for each pixel constituting the fusion block; As the absolute value of the difference between the pixel value of the pixel belonging to the target block and the pixel value of the corresponding pixel on the overlap target block or the destination block increases, the overlap target block gradually inherits to the fusion block. An image processing apparatus, wherein the weighting of superposition is changed so as not to occur. The image processing apparatus according to claim 1 or 2,
The superposition means is configured to superimpose the target block on the original image and the superposition target block on the superposition target image with individual weights every time the fusion block is generated, and on the target block As the absolute value of the difference between the pixel value of the pixel and the pixel value on the overlap target block or the destination block increases, the overlapping weight is gradually changed so that the overlap target block is not inherited by the fusion block. A featured image processing apparatus. The image processing apparatus according to any one of claims 1 to 4,
When the same position block at the position of the target block is set in the reference image and the destination block is not similar to the target block as compared with the same position block, the search means has searched An image processing apparatus, comprising: an editing unit that performs an overwriting on the output of the search unit so that a destination block is the same position block. The image processing apparatus according to any one of claims 1 to 5,
(A1) The superimposition target image is a noise reduction image obtained when image processing is performed on an image photographed before the original image, and (b1) the reference image is positioned before the original image. An image processing apparatus, which is a captured image. The image processing apparatus according to any one of claims 1 to 5,
(A1) The superimposition target image is a noise-reduced image obtained when image processing is performed on an image captured before the original image. (B2) The reference image is positioned before the original image. An image processing apparatus, wherein the image processing apparatus is a noise-reduced image corresponding to a photographed image. The image processing apparatus according to any one of claims 1 to 5,
(A2) The image to be superposed is an image taken before the original image, and (b1) the reference image is an image taken before the original image. . The image processing apparatus according to any one of claims 1 to 5,
(A2) The image to be overlapped is an image taken before the original image, and (b2) the reference image is a noise-reduced image corresponding to an image taken before the original image. A featured image processing apparatus. The image processing apparatus according to any one of claims 1 to 9,
The search means is an accuracy priority mode for searching a destination block for a certain target pixel over a wide range of the reference image, and
A speed priority mode for searching for a destination block for a target pixel different from the target pixel that has been processed in the accuracy priority mode based on a search result of the accuracy priority mode over a narrow range of the reference image Works based on two modes of
The speed priority mode is based on the positional relationship between the target pixel that is the processing target of the accuracy priority mode and the destination pixel of the target pixel found in the search of the accuracy priority mode on the reference image. An image processing characterized in that, based on the reference image, the position of the target pixel of the target pixel currently being searched is predicted, and the target block is searched over a range surrounding the predicted position. apparatus. The image processing apparatus according to any one of claims 1 to 10,
An image processing apparatus according to claim 1, wherein a range of the block to be overlapped according to the overlapping unit is narrower than a range of the target block and the destination block. The image processing apparatus according to any one of claims 1 to 10,
An image processing apparatus according to claim 1, wherein a range of the block to be overlapped by the overlapping unit is wider than a range of the target block and the destination block. The image processing apparatus according to any one of claims 1 to 12,
Image reduction means for reducing the original image and the reference image to generate a reduced original image and a reduced reference image;
Reduced image target block setting means for setting a target block on the reduced original image that is the target block on the target pixel and the reduced original image from among the pixels constituting the reduced original image;
Reduced image search means for searching a reduced reference image destination block most similar to the reduced original image target block from the reduced reference image;
Search range setting means for setting a search range in which the search means searches for the destination block in the reference image based on the position on the reference image corresponding to the position of the destination block on the reduced reference image. An image processing apparatus comprising: The image processing apparatus according to any one of claims 1 to 13,
When the search means searches for the destination block from candidate blocks that are candidates for the destination block in the reference image, the search means also determines similarity of the candidate blocks that have been rotated. An image processing apparatus. The image processing apparatus according to any one of claims 1 to 14,
Target block according to the target setting means is an image processing apparatus characterized by having an enclave. The image processing apparatus according to any one of claims 1 to 15,
The image processing apparatus wherein the target setting target block according to the constant means, characterized in that it is configured to exclude a portion of the peripheral pixels surrounding the pixel of interest. The image processing apparatus according to any one of claims 1 to 16,
The search means is configured to search for a destination block from each of a plurality of different reference images,
The superposition means generates the fusion block by superimposing each superposition target block corresponding to each of the destination blocks searched in each of the reference images on the target block on the original image. An image processing apparatus. The image processing apparatus according to any one of claims 1 to 16,
The search means is configured to search for a destination block from each of a plurality of different reference images,
The overlapping unit generates a plurality of fusion blocks by superimposing the target block on the original image on each of the overlapping blocks corresponding to each of the destination blocks searched for in each of the reference images. An image processing apparatus that generates a final fusion block by superimposing blocks on each other. The image processing device according to any one of claims 1 to 18,
Based on the positional relationship between the target block and the destination block output by the search means, the source image is obtained by grasping where the target pixel for each of the target blocks has moved on the reference image. Recognize the movement destination on the reference image for each of the constituting pixels, and move the pixel when the movement direction and movement distance of a certain pixel on the original image are far from the surrounding pixels of the pixel. An image processing apparatus, comprising: an editing unit that performs editing to change the destination block corresponding to the target block so that the destination is a position moved by the same distance in the same direction as the movement of surrounding pixels. The image processing apparatus according to any one of claims 1 to 16,
The search means is configured to search for a plurality of destination blocks from the reference image,
The overlapping unit generates a plurality of fusion blocks by superimposing the target block on the original image on each of the overlapping blocks corresponding to each of the plurality of destination blocks searched for in the reference image, An image processing apparatus that generates a final fusion block by superimposing blocks on each other. The image processing apparatus according to any one of claims 1 to 20,
When searching for a destination block from candidate blocks that are candidates for a destination block in the reference image, the search means selects a candidate block close to the position on the reference image corresponding to the target block in the original image. An image processing apparatus that preferentially recognizes the destination block. The image processing apparatus according to any one of claims 1 to 21,
The image generation means, based on the degree of variation indicating the degree of difference in the pattern reflected in the target block that is the source of the fusion block and the destination block, the fusion block with the higher degree of variation is included in the noise-reduced image. An image processing apparatus that generates the noise-reduced image by overlapping and adding fusion blocks while weighting so as to be thinly reflected. The image processing apparatus according to any one of claims 1 to 22,
The target processing unit operates by distinguishing a pixel on the original image into a pixel that is set as the target pixel and a pixel that is not set.

Images